id	sid	tid	token	lemma	pos
bracis-19021	1	1	gradient	gradient	NOUN
bracis-19021	1	2	estimation	estimation	NOUN
bracis-19021	1	3	in	in	ADP
bracis-19021	1	4	model	model	NOUN
bracis-19021	1	5	-	-	PUNCT
bracis-19021	1	6	based	base	VERB
bracis-19021	1	7	reinforcement	reinforcement	NOUN
bracis-19021	1	8	learning	learning	NOUN
bracis-19021	1	9	:	:	PUNCT
bracis-19021	1	10	a	a	DET
bracis-19021	1	11	study	study	NOUN
bracis-19021	1	12	on	on	ADP
bracis-19021	1	13	linear	linear	ADJ
bracis-19021	1	14	quadratic	quadratic	ADJ
bracis-19021	1	15	environments	environment	NOUN
bracis-19021	1	16	|	|	ADV
bracis-19021	1	17	springer	springer	NOUN
bracis-19021	1	18	nature	nature	PROPN
bracis-19021	1	19	link	link	PROPN
bracis-19021	1	20	(	(	PUNCT
bracis-19021	1	21	formerly	formerly	ADV
bracis-19021	1	22	springerlink	springerlink	NOUN
bracis-19021	1	23	)	)	PUNCT
bracis-19021	1	24	skip	skip	VERB
bracis-19021	1	25	to	to	ADP
bracis-19021	1	26	main	main	ADJ
bracis-19021	1	27	content	content	NOUN
bracis-19021	1	28	advertisement	advertisement	NOUN
bracis-19021	1	29	log	log	NOUN
bracis-19021	1	30	in	in	ADP
bracis-19021	1	31	menu	menu	NOUN
bracis-19021	1	32	find	find	VERB
bracis-19021	1	33	a	a	DET
bracis-19021	1	34	journal	journal	NOUN
bracis-19021	1	35	publish	publish	VERB
bracis-19021	1	36	with	with	ADP
bracis-19021	1	37	us	we	PRON
bracis-19021	1	38	track	track	VERB
bracis-19021	1	39	your	your	PRON
bracis-19021	1	40	research	research	NOUN
bracis-19021	1	41	search	search	NOUN
bracis-19021	1	42	cart	cart	NOUN
bracis-19021	1	43	home	home	NOUN
bracis-19021	2	1	intelligent	intelligent	ADJ
bracis-19021	2	2	systems	system	NOUN
bracis-19021	2	3	conference	conference	NOUN
bracis-19021	2	4	paper	paper	NOUN
bracis-19021	2	5	gradient	gradient	NOUN
bracis-19021	2	6	estimation	estimation	NOUN
bracis-19021	2	7	in	in	ADP
bracis-19021	2	8	model	model	NOUN
bracis-19021	2	9	-	-	PUNCT
bracis-19021	2	10	based	base	VERB
bracis-19021	2	11	reinforcement	reinforcement	NOUN
bracis-19021	2	12	learning	learning	NOUN
bracis-19021	2	13	:	:	PUNCT
bracis-19021	2	14	a	a	DET
bracis-19021	2	15	study	study	NOUN
bracis-19021	2	16	on	on	ADP
bracis-19021	2	17	linear	linear	ADJ
bracis-19021	2	18	quadratic	quadratic	ADJ
bracis-19021	2	19	environments	environment	NOUN
bracis-19021	2	20	conference	conference	NOUN
bracis-19021	2	21	paper	paper	NOUN
bracis-19021	2	22	first	first	ADV
bracis-19021	2	23	online	online	ADV
bracis-19021	2	24	:	:	PUNCT
bracis-19021	2	25	28	28	NUM
bracis-19021	2	26	november	november	PROPN
bracis-19021	2	27	2021	2021	NUM
bracis-19021	2	28	pp	pp	ADP
bracis-19021	2	29	33–47	33–47	NUM
bracis-19021	2	30	cite	cite	VERB
bracis-19021	2	31	this	this	DET
bracis-19021	2	32	conference	conference	NOUN
bracis-19021	2	33	paper	paper	NOUN
bracis-19021	2	34	access	access	NOUN
bracis-19021	2	35	provided	provide	VERB
bracis-19021	2	36	by	by	ADP
bracis-19021	2	37	university	university	PROPN
bracis-19021	2	38	of	of	ADP
bracis-19021	2	39	notre	notre	PROPN
bracis-19021	2	40	dame	dame	PROPN
bracis-19021	2	41	hesburgh	hesburgh	PROPN
bracis-19021	2	42	library	library	PROPN
bracis-19021	2	43	download	download	PROPN
bracis-19021	2	44	book	book	NOUN
bracis-19021	2	45	pdf	pdf	PROPN
bracis-19021	2	46	download	download	NOUN
bracis-19021	2	47	book	book	NOUN
bracis-19021	2	48	epub	epub	PROPN
bracis-19021	2	49	intelligent	intelligent	ADJ
bracis-19021	2	50	systems	system	NOUN
bracis-19021	2	51	(	(	PUNCT
bracis-19021	2	52	bracis	bracis	NOUN
bracis-19021	2	53	2021	2021	NUM
bracis-19021	2	54	)	)	PUNCT
bracis-19021	2	55	gradient	gradient	NOUN
bracis-19021	2	56	estimation	estimation	NOUN
bracis-19021	2	57	in	in	ADP
bracis-19021	2	58	model	model	NOUN
bracis-19021	2	59	-	-	PUNCT
bracis-19021	2	60	based	base	VERB
bracis-19021	2	61	reinforcement	reinforcement	NOUN
bracis-19021	2	62	learning	learning	NOUN
bracis-19021	2	63	:	:	PUNCT
bracis-19021	2	64	a	a	DET
bracis-19021	2	65	study	study	NOUN
bracis-19021	2	66	on	on	ADP
bracis-19021	2	67	linear	linear	ADJ
bracis-19021	2	68	quadratic	quadratic	ADJ
bracis-19021	2	69	environments	environment	NOUN
bracis-19021	2	70	download	download	NOUN
bracis-19021	2	71	book	book	NOUN
bracis-19021	2	72	pdf	pdf	PROPN
bracis-19021	2	73	download	download	NOUN
bracis-19021	2	74	book	book	NOUN
bracis-19021	2	75	epub	epub	PROPN
bracis-19021	2	76	ângelo	ângelo	VERB
bracis-19021	2	77	gregório	gregório	PROPN
bracis-19021	2	78	lovatto	lovatto	PROPN
bracis-19021	2	79	  	  	SPACE
bracis-19021	2	80	orcid	orcid	PROPN
bracis-19021	2	81	:	:	PUNCT
bracis-19021	2	82	orcid.org/0000-0002-0586-541210	orcid.org/0000-0002-0586-541210	PROPN
bracis-19021	2	83	,	,	PUNCT
bracis-19021	2	84	thiago	thiago	PROPN
bracis-19021	2	85	pereira	pereira	PROPN
bracis-19021	2	86	bueno	bueno	PROPN
bracis-19021	2	87	  	  	SPACE
bracis-19021	2	88	orcid	orcid	PROPN
bracis-19021	2	89	:	:	PUNCT
bracis-19021	2	90	orcid.org/0000-0003-1999-969710	orcid.org/0000-0003-1999-969710	PROPN
bracis-19021	2	91	&	&	CCONJ
bracis-19021	2	92	leliane	leliane	PROPN
bracis-19021	2	93	nunes	nunes	PROPN
bracis-19021	2	94	de	de	PROPN
bracis-19021	2	95	barros	barros	PROPN
bracis-19021	2	96	  	  	SPACE
bracis-19021	2	97	orcid	orcid	NOUN
bracis-19021	2	98	:	:	PUNCT
bracis-19021	2	99	orcid.org/0000-0002-9128-018810	orcid.org/0000-0002-9128-018810	NOUN
bracis-19021	2	100	  	  	SPACE
bracis-19021	2	101	part	part	NOUN
bracis-19021	2	102	of	of	ADP
bracis-19021	2	103	the	the	DET
bracis-19021	2	104	book	book	NOUN
bracis-19021	2	105	series	series	NOUN
bracis-19021	2	106	:	:	PUNCT
bracis-19021	2	107	lecture	lecture	NOUN
bracis-19021	2	108	notes	note	NOUN
bracis-19021	2	109	in	in	ADP
bracis-19021	2	110	computer	computer	NOUN
bracis-19021	2	111	science	science	NOUN
bracis-19021	2	112	(	(	PUNCT
bracis-19021	2	113	(	(	PUNCT
bracis-19021	2	114	lnai	lnai	ADJ
bracis-19021	2	115	,	,	PUNCT
bracis-19021	2	116	volume	volume	NOUN
bracis-19021	2	117	13073	13073	NUM
bracis-19021	2	118	)	)	PUNCT
bracis-19021	2	119	)	)	PUNCT
bracis-19021	2	120	included	include	VERB
bracis-19021	2	121	in	in	ADP
bracis-19021	2	122	the	the	DET
bracis-19021	2	123	following	follow	VERB
bracis-19021	2	124	conference	conference	NOUN
bracis-19021	2	125	series	series	NOUN
bracis-19021	2	126	:	:	PUNCT
bracis-19021	2	127	brazilian	brazilian	ADJ
bracis-19021	2	128	conference	conference	NOUN
bracis-19021	2	129	on	on	ADP
bracis-19021	2	130	intelligent	intelligent	ADJ
bracis-19021	2	131	systems	system	NOUN
bracis-19021	2	132	772	772	NUM
bracis-19021	2	133	accesses	access	NOUN
bracis-19021	2	134	1	1	NUM
bracis-19021	2	135	citation	citation	NOUN
bracis-19021	2	136	abstract	abstract	ADJ
bracis-19021	2	137	stochastic	stochastic	NOUN
bracis-19021	2	138	value	value	NOUN
bracis-19021	2	139	gradient	gradient	NOUN
bracis-19021	2	140	(	(	PUNCT
bracis-19021	2	141	svg	svg	NOUN
bracis-19021	2	142	)	)	PUNCT
bracis-19021	2	143	methods	method	NOUN
bracis-19021	2	144	underlie	underlie	VERB
bracis-19021	2	145	many	many	ADJ
bracis-19021	2	146	recent	recent	ADJ
bracis-19021	2	147	achievements	achievement	NOUN
bracis-19021	2	148	of	of	ADP
bracis-19021	2	149	model	model	NOUN
bracis-19021	2	150	-	-	PUNCT
bracis-19021	2	151	based	base	VERB
bracis-19021	2	152	reinforcement	reinforcement	NOUN
bracis-19021	2	153	learning	learning	NOUN
bracis-19021	2	154	agents	agent	NOUN
bracis-19021	2	155	in	in	ADP
bracis-19021	2	156	continuous	continuous	ADJ
bracis-19021	2	157	state	state	NOUN
bracis-19021	2	158	-	-	PUNCT
bracis-19021	2	159	action	action	NOUN
bracis-19021	2	160	spaces	space	NOUN
bracis-19021	2	161	.	.	PUNCT
bracis-19021	3	1	despite	despite	SCONJ
bracis-19021	3	2	their	their	PRON
bracis-19021	3	3	practical	practical	ADJ
bracis-19021	3	4	significance	significance	NOUN
bracis-19021	3	5	,	,	PUNCT
bracis-19021	3	6	many	many	ADJ
bracis-19021	3	7	algorithm	algorithm	NOUN
bracis-19021	3	8	design	design	NOUN
bracis-19021	3	9	choices	choice	NOUN
bracis-19021	3	10	still	still	ADV
bracis-19021	3	11	lack	lack	VERB
bracis-19021	3	12	rigorous	rigorous	ADJ
bracis-19021	3	13	theoretical	theoretical	ADJ
bracis-19021	3	14	or	or	CCONJ
bracis-19021	3	15	empirical	empirical	ADJ
bracis-19021	3	16	justification	justification	NOUN
bracis-19021	3	17	.	.	PUNCT
bracis-19021	4	1	in	in	ADP
bracis-19021	4	2	this	this	DET
bracis-19021	4	3	work	work	NOUN
bracis-19021	4	4	,	,	PUNCT
bracis-19021	4	5	we	we	PRON
bracis-19021	4	6	analyze	analyze	VERB
bracis-19021	4	7	one	one	NUM
bracis-19021	4	8	such	such	ADJ
bracis-19021	4	9	design	design	NOUN
bracis-19021	4	10	choice	choice	NOUN
bracis-19021	4	11	:	:	PUNCT
bracis-19021	4	12	the	the	DET
bracis-19021	4	13	gradient	gradient	ADJ
bracis-19021	4	14	estimator	estimator	NOUN
bracis-19021	4	15	formula	formula	NOUN
bracis-19021	4	16	.	.	PUNCT
bracis-19021	5	1	we	we	PRON
bracis-19021	5	2	conduct	conduct	VERB
bracis-19021	5	3	our	our	PRON
bracis-19021	5	4	analysis	analysis	NOUN
bracis-19021	5	5	on	on	ADP
bracis-19021	5	6	randomized	randomized	ADJ
bracis-19021	5	7	linear	linear	ADJ
bracis-19021	5	8	quadratic	quadratic	ADJ
bracis-19021	5	9	gaussian	gaussian	ADJ
bracis-19021	5	10	environments	environment	NOUN
bracis-19021	5	11	,	,	PUNCT
bracis-19021	5	12	allowing	allow	VERB
bracis-19021	5	13	us	we	PRON
bracis-19021	5	14	to	to	PART
bracis-19021	5	15	empirically	empirically	ADV
bracis-19021	5	16	assess	assess	VERB
bracis-19021	5	17	gradient	gradient	ADJ
bracis-19021	5	18	estimation	estimation	NOUN
bracis-19021	5	19	quality	quality	NOUN
bracis-19021	5	20	relative	relative	ADJ
bracis-19021	5	21	to	to	ADP
bracis-19021	5	22	the	the	DET
bracis-19021	5	23	actual	actual	ADJ
bracis-19021	5	24	svg	svg	PROPN
bracis-19021	5	25	.	.	PUNCT
bracis-19021	6	1	our	our	PRON
bracis-19021	6	2	results	result	NOUN
bracis-19021	6	3	justify	justify	VERB
bracis-19021	6	4	a	a	DET
bracis-19021	6	5	widely	widely	ADV
bracis-19021	6	6	used	use	VERB
bracis-19021	6	7	gradient	gradient	NOUN
bracis-19021	6	8	estimator	estimator	NOUN
bracis-19021	6	9	by	by	ADP
bracis-19021	6	10	showing	show	VERB
bracis-19021	6	11	it	it	PRON
bracis-19021	6	12	induces	induce	VERB
bracis-19021	6	13	a	a	DET
bracis-19021	6	14	favorable	favorable	ADJ
bracis-19021	6	15	bias	bias	NOUN
bracis-19021	6	16	-	-	PUNCT
bracis-19021	6	17	variance	variance	NOUN
bracis-19021	6	18	tradeoff	tradeoff	NOUN
bracis-19021	6	19	,	,	PUNCT
bracis-19021	6	20	which	which	PRON
bracis-19021	6	21	could	could	AUX
bracis-19021	6	22	explain	explain	VERB
bracis-19021	6	23	the	the	DET
bracis-19021	6	24	lower	low	ADJ
bracis-19021	6	25	sample	sample	NOUN
bracis-19021	6	26	complexity	complexity	NOUN
bracis-19021	6	27	of	of	ADP
bracis-19021	6	28	recent	recent	ADJ
bracis-19021	6	29	svg	svg	NOUN
bracis-19021	6	30	methods	method	NOUN
bracis-19021	6	31	.	.	PUNCT
bracis-19021	7	1	access	access	NOUN
bracis-19021	7	2	provided	provide	VERB
bracis-19021	7	3	by	by	ADP
bracis-19021	7	4	university	university	PROPN
bracis-19021	7	5	of	of	ADP
bracis-19021	7	6	notre	notre	PROPN
bracis-19021	7	7	dame	dame	PROPN
bracis-19021	7	8	hesburgh	hesburgh	PROPN
bracis-19021	7	9	library	library	PROPN
bracis-19021	7	10	.	.	PUNCT
bracis-19021	8	1	download	download	PROPN
bracis-19021	8	2	conference	conference	NOUN
bracis-19021	8	3	paper	paper	NOUN
bracis-19021	8	4	pdf	pdf	NOUN
bracis-19021	8	5	similar	similar	ADJ
bracis-19021	8	6	content	content	NOUN
bracis-19021	8	7	being	be	AUX
bracis-19021	8	8	viewed	view	VERB
bracis-19021	8	9	by	by	ADP
bracis-19021	8	10	others	other	NOUN
bracis-19021	8	11	exploration	exploration	NOUN
bracis-19021	8	12	versus	versus	ADP
bracis-19021	8	13	exploitation	exploitation	NOUN
bracis-19021	8	14	in	in	ADP
bracis-19021	8	15	 	 	SPACE
bracis-19021	8	16	model	model	NOUN
bracis-19021	8	17	-	-	PUNCT
bracis-19021	8	18	based	base	VERB
bracis-19021	8	19	reinforcement	reinforcement	NOUN
bracis-19021	8	20	learning	learning	NOUN
bracis-19021	8	21	:	:	PUNCT
bracis-19021	8	22	an	an	DET
bracis-19021	8	23	empirical	empirical	ADJ
bracis-19021	8	24	study	study	NOUN
bracis-19021	8	25	chapter	chapter	NOUN
bracis-19021	8	26	©	©	PROPN
bracis-19021	8	27	2022	2022	NUM
bracis-19021	8	28	randomized	randomize	VERB
bracis-19021	8	29	smoothing	smoothing	NOUN
bracis-19021	8	30	variance	variance	NOUN
bracis-19021	8	31	reduction	reduction	NOUN
bracis-19021	8	32	method	method	NOUN
bracis-19021	8	33	for	for	ADP
bracis-19021	8	34	large	large	ADJ
bracis-19021	8	35	-	-	PUNCT
bracis-19021	8	36	scale	scale	NOUN
bracis-19021	8	37	non	non	ADJ
bracis-19021	8	38	-	-	ADJ
bracis-19021	8	39	smooth	smooth	ADJ
bracis-19021	8	40	convex	convex	NOUN
bracis-19021	8	41	optimization	optimization	NOUN
bracis-19021	8	42	article	article	NOUN
bracis-19021	8	43	24	24	NUM
bracis-19021	8	44	may	may	PROPN
bracis-19021	8	45	2021	2021	NUM
bracis-19021	8	46	superquantile	superquantile	NOUN
bracis-19021	8	47	-	-	PUNCT
bracis-19021	8	48	based	base	VERB
bracis-19021	8	49	learning	learning	NOUN
bracis-19021	8	50	:	:	PUNCT
bracis-19021	8	51	a	a	DET
bracis-19021	8	52	direct	direct	ADJ
bracis-19021	8	53	approach	approach	NOUN
bracis-19021	8	54	using	use	VERB
bracis-19021	8	55	gradient	gradient	NOUN
bracis-19021	8	56	-	-	PUNCT
bracis-19021	8	57	based	base	VERB
bracis-19021	8	58	optimization	optimization	NOUN
bracis-19021	8	59	article	article	NOUN
bracis-19021	8	60	11	11	NUM
bracis-19021	8	61	january	january	PROPN
bracis-19021	8	62	2022	2022	NUM
bracis-19021	8	63	explore	explore	VERB
bracis-19021	8	64	related	relate	VERB
bracis-19021	8	65	subjects	subject	NOUN
bracis-19021	8	66	discover	discover	VERB
bracis-19021	8	67	the	the	DET
bracis-19021	8	68	latest	late	ADJ
bracis-19021	8	69	articles	article	NOUN
bracis-19021	8	70	,	,	PUNCT
bracis-19021	8	71	books	book	NOUN
bracis-19021	8	72	and	and	CCONJ
bracis-19021	8	73	news	news	NOUN
bracis-19021	8	74	in	in	ADP
bracis-19021	8	75	related	related	ADJ
bracis-19021	8	76	subjects	subject	NOUN
bracis-19021	8	77	,	,	PUNCT
bracis-19021	8	78	suggested	suggest	VERB
bracis-19021	8	79	using	use	VERB
bracis-19021	8	80	machine	machine	NOUN
bracis-19021	8	81	learning	learning	NOUN
bracis-19021	8	82	.	.	PUNCT
bracis-19021	9	1	learning	learn	VERB
bracis-19021	9	2	algorithms	algorithm	NOUN
bracis-19021	9	3	linear	linear	PROPN
bracis-19021	9	4	algebra	algebra	NOUN
bracis-19021	9	5	signalling	signal	VERB
bracis-19021	9	6	gradients	gradient	NOUN
bracis-19021	9	7	statistical	statistical	ADJ
bracis-19021	9	8	learning	learning	NOUN
bracis-19021	9	9	stochastic	stochastic	NOUN
bracis-19021	9	10	learning	learning	NOUN
bracis-19021	9	11	and	and	CCONJ
bracis-19021	9	12	adaptive	adaptive	ADJ
bracis-19021	9	13	control	control	NOUN
bracis-19021	9	14	calculus	calculus	NOUN
bracis-19021	9	15	of	of	ADP
bracis-19021	9	16	variations	variation	NOUN
bracis-19021	9	17	and	and	CCONJ
bracis-19021	9	18	optimization	optimization	NOUN
bracis-19021	9	19	1	1	NUM
bracis-19021	9	20	introduction	introduction	NOUN
bracis-19021	9	21	model	model	NOUN
bracis-19021	9	22	-	-	PUNCT
bracis-19021	9	23	based	base	VERB
bracis-19021	9	24	reinforcement	reinforcement	NOUN
bracis-19021	9	25	learning	learning	NOUN
bracis-19021	9	26	(	(	PUNCT
bracis-19021	9	27	mbrl	mbrl	PROPN
bracis-19021	9	28	)	)	PUNCT
bracis-19021	10	1	[	[	X
bracis-19021	10	2	16	16	NUM
bracis-19021	10	3	,	,	PUNCT
bracis-19021	10	4	20	20	NUM
bracis-19021	10	5	]	]	PUNCT
bracis-19021	10	6	is	be	AUX
bracis-19021	10	7	a	a	DET
bracis-19021	10	8	promising	promising	ADJ
bracis-19021	10	9	framework	framework	NOUN
bracis-19021	10	10	for	for	ADP
bracis-19021	10	11	developing	develop	VERB
bracis-19021	10	12	intelligent	intelligent	ADJ
bracis-19021	10	13	systems	system	NOUN
bracis-19021	10	14	for	for	ADP
bracis-19021	10	15	sequential	sequential	ADJ
bracis-19021	10	16	decision	decision	NOUN
bracis-19021	10	17	-	-	PUNCT
bracis-19021	10	18	making	making	NOUN
bracis-19021	10	19	from	from	ADP
bracis-19021	10	20	limited	limited	ADJ
bracis-19021	10	21	data	datum	NOUN
bracis-19021	10	22	.	.	PUNCT
bracis-19021	11	1	unlike	unlike	ADP
bracis-19021	11	2	in	in	ADP
bracis-19021	11	3	model	model	ADJ
bracis-19021	11	4	-	-	PUNCT
bracis-19021	11	5	free	free	ADJ
bracis-19021	11	6	reinforcement	reinforcement	NOUN
bracis-19021	11	7	learning	learning	NOUN
bracis-19021	11	8	(	(	PUNCT
bracis-19021	11	9	rl	rl	NOUN
bracis-19021	11	10	)	)	PUNCT
bracis-19021	11	11	methods	method	NOUN
bracis-19021	11	12	,	,	PUNCT
bracis-19021	11	13	mbrl	mbrl	PROPN
bracis-19021	11	14	agents	agent	NOUN
bracis-19021	11	15	use	use	VERB
bracis-19021	11	16	collected	collect	VERB
bracis-19021	11	17	experiences	experience	NOUN
bracis-19021	11	18	to	to	PART
bracis-19021	11	19	fit	fit	VERB
bracis-19021	11	20	a	a	DET
bracis-19021	11	21	predictive	predictive	ADJ
bracis-19021	11	22	model	model	NOUN
bracis-19021	11	23	of	of	ADP
bracis-19021	11	24	the	the	DET
bracis-19021	11	25	environment	environment	NOUN
bracis-19021	11	26	.	.	PUNCT
bracis-19021	12	1	the	the	DET
bracis-19021	12	2	agent	agent	NOUN
bracis-19021	12	3	can	can	AUX
bracis-19021	12	4	use	use	VERB
bracis-19021	12	5	the	the	DET
bracis-19021	12	6	model	model	NOUN
bracis-19021	12	7	to	to	PART
bracis-19021	12	8	evaluate	evaluate	VERB
bracis-19021	12	9	potential	potential	ADJ
bracis-19021	12	10	action	action	NOUN
bracis-19021	12	11	sequences	sequence	NOUN
bracis-19021	12	12	,	,	PUNCT
bracis-19021	12	13	saving	save	VERB
bracis-19021	12	14	costly	costly	ADJ
bracis-19021	12	15	trial	trial	NOUN
bracis-19021	12	16	-	-	PUNCT
bracis-19021	12	17	and	and	CCONJ
bracis-19021	12	18	-	-	PUNCT
bracis-19021	12	19	error	error	NOUN
bracis-19021	12	20	experimentation	experimentation	NOUN
bracis-19021	12	21	in	in	ADP
bracis-19021	12	22	the	the	DET
bracis-19021	12	23	real	real	ADJ
bracis-19021	12	24	world	world	NOUN
bracis-19021	12	25	,	,	PUNCT
bracis-19021	12	26	or	or	CCONJ
bracis-19021	12	27	to	to	PART
bracis-19021	12	28	estimate	estimate	VERB
bracis-19021	12	29	quantities	quantity	NOUN
bracis-19021	12	30	useful	useful	ADJ
bracis-19021	12	31	for	for	ADP
bracis-19021	12	32	improving	improve	VERB
bracis-19021	12	33	its	its	PRON
bracis-19021	12	34	learned	learn	VERB
bracis-19021	12	35	behavior	behavior	NOUN
bracis-19021	12	36	.	.	PUNCT
bracis-19021	13	1	stochastic	stochastic	ADJ
bracis-19021	13	2	value	value	NOUN
bracis-19021	13	3	gradient	gradient	NOUN
bracis-19021	13	4	(	(	PUNCT
bracis-19021	13	5	svg	svg	NOUN
bracis-19021	13	6	)	)	PUNCT
bracis-19021	13	7	methods	method	NOUN
bracis-19021	13	8	belong	belong	VERB
bracis-19021	13	9	to	to	ADP
bracis-19021	13	10	the	the	DET
bracis-19021	13	11	latter	latter	ADJ
bracis-19021	13	12	category	category	NOUN
bracis-19021	13	13	,	,	PUNCT
bracis-19021	13	14	using	use	VERB
bracis-19021	13	15	the	the	DET
bracis-19021	13	16	model	model	NOUN
bracis-19021	13	17	to	to	PART
bracis-19021	13	18	estimate	estimate	VERB
bracis-19021	13	19	the	the	DET
bracis-19021	13	20	value	value	NOUN
bracis-19021	13	21	gradient	gradient	NOUN
bracis-19021	13	22	.	.	PUNCT
bracis-19021	14	1	while	while	SCONJ
bracis-19021	14	2	model	model	ADJ
bracis-19021	14	3	-	-	PUNCT
bracis-19021	14	4	free	free	ADJ
bracis-19021	14	5	methods	method	NOUN
bracis-19021	14	6	use	use	VERB
bracis-19021	14	7	the	the	DET
bracis-19021	14	8	score	score	NOUN
bracis-19021	14	9	-	-	PUNCT
bracis-19021	14	10	function	function	NOUN
bracis-19021	14	11	estimator	estimator	NOUN
bracis-19021	14	12	of	of	ADP
bracis-19021	14	13	the	the	DET
bracis-19021	14	14	value	value	NOUN
bracis-19021	14	15	gradient	gradient	NOUN
bracis-19021	14	16	,	,	PUNCT
bracis-19021	14	17	svg	svg	PROPN
bracis-19021	14	18	methods	method	NOUN
bracis-19021	14	19	can	can	AUX
bracis-19021	14	20	leverage	leverage	VERB
bracis-19021	14	21	the	the	DET
bracis-19021	14	22	model	model	NOUN
bracis-19021	14	23	to	to	PART
bracis-19021	14	24	produce	produce	VERB
bracis-19021	14	25	gradients	gradient	NOUN
bracis-19021	14	26	via	via	ADP
bracis-19021	14	27	the	the	DET
bracis-19021	14	28	pathwise	pathwise	NOUN
bracis-19021	14	29	derivative	derivative	ADJ
bracis-19021	14	30	estimator	estimator	NOUN
bracis-19021	14	31	,	,	PUNCT
bracis-19021	14	32	usually	usually	ADV
bracis-19021	14	33	found	find	VERB
bracis-19021	14	34	to	to	PART
bracis-19021	14	35	be	be	AUX
bracis-19021	14	36	more	more	ADV
bracis-19021	14	37	stable	stable	ADJ
bracis-19021	14	38	in	in	ADP
bracis-19021	14	39	practice	practice	NOUN
bracis-19021	14	40	[	[	X
bracis-19021	14	41	23	23	NUM
bracis-19021	14	42	]	]	PUNCT
bracis-19021	14	43	.	.	PUNCT
bracis-19021	15	1	recently	recently	ADV
bracis-19021	15	2	proposed	propose	VERB
bracis-19021	15	3	rl	rl	ADP
bracis-19021	15	4	agents	agent	NOUN
bracis-19021	15	5	using	use	VERB
bracis-19021	15	6	the	the	DET
bracis-19021	15	7	svg	svg	NOUN
bracis-19021	15	8	approach	approach	NOUN
bracis-19021	15	9	have	have	AUX
bracis-19021	15	10	demonstrated	demonstrate	VERB
bracis-19021	15	11	its	its	PRON
bracis-19021	15	12	effectiveness	effectiveness	NOUN
bracis-19021	15	13	in	in	ADP
bracis-19021	15	14	learning	learn	VERB
bracis-19021	15	15	robotic	robotic	ADJ
bracis-19021	15	16	locomotion	locomotion	NOUN
bracis-19021	15	17	from	from	ADP
bracis-19021	15	18	data	datum	NOUN
bracis-19021	15	19	with	with	ADP
bracis-19021	15	20	unprecedented	unprecedented	ADJ
bracis-19021	15	21	sample	sample	NOUN
bracis-19021	15	22	-	-	PUNCT
bracis-19021	15	23	efficiency	efficiency	NOUN
bracis-19021	15	24	[	[	X
bracis-19021	15	25	1	1	NUM
bracis-19021	15	26	,	,	PUNCT
bracis-19021	15	27	4	4	NUM
bracis-19021	15	28	,	,	PUNCT
bracis-19021	15	29	8	8	NUM
bracis-19021	15	30	]	]	PUNCT
bracis-19021	15	31	,	,	PUNCT
bracis-19021	15	32	i.e.	i.e.	X
bracis-19021	15	33	with	with	ADP
bracis-19021	15	34	few	few	ADJ
bracis-19021	15	35	collected	collect	VERB
bracis-19021	15	36	experiences	experience	NOUN
bracis-19021	15	37	.	.	PUNCT
bracis-19021	16	1	although	although	SCONJ
bracis-19021	16	2	svg	svg	PROPN
bracis-19021	16	3	methods	method	NOUN
bracis-19021	16	4	have	have	AUX
bracis-19021	16	5	been	be	AUX
bracis-19021	16	6	validated	validate	VERB
bracis-19021	16	7	empirically	empirically	ADV
bracis-19021	16	8	,	,	PUNCT
bracis-19021	16	9	some	some	PRON
bracis-19021	16	10	of	of	ADP
bracis-19021	16	11	the	the	DET
bracis-19021	16	12	algorithm	algorithm	NOUN
bracis-19021	16	13	design	design	NOUN
bracis-19021	16	14	choices	choice	NOUN
bracis-19021	16	15	still	still	ADV
bracis-19021	16	16	lack	lack	VERB
bracis-19021	16	17	rigorous	rigorous	ADJ
bracis-19021	16	18	theoretical	theoretical	ADJ
bracis-19021	16	19	or	or	CCONJ
bracis-19021	16	20	empirical	empirical	ADJ
bracis-19021	16	21	justification	justification	NOUN
bracis-19021	16	22	.	.	PUNCT
bracis-19021	17	1	as	as	SCONJ
bracis-19021	17	2	recent	recent	ADJ
bracis-19021	17	3	work	work	NOUN
bracis-19021	17	4	on	on	ADP
bracis-19021	17	5	model	model	ADJ
bracis-19021	17	6	-	-	PUNCT
bracis-19021	17	7	free	free	ADJ
bracis-19021	17	8	methods	method	NOUN
bracis-19021	17	9	has	have	AUX
bracis-19021	17	10	shown	show	VERB
bracis-19021	17	11	,	,	PUNCT
bracis-19021	17	12	there	there	PRON
bracis-19021	17	13	can	can	AUX
bracis-19021	17	14	be	be	AUX
bracis-19021	17	15	a	a	DET
bracis-19021	17	16	gap	gap	NOUN
bracis-19021	17	17	between	between	ADP
bracis-19021	17	18	the	the	DET
bracis-19021	17	19	theoretical	theoretical	ADJ
bracis-19021	17	20	underpinnings	underpinning	NOUN
bracis-19021	17	21	of	of	ADP
bracis-19021	17	22	rl	rl	NOUN
bracis-19021	17	23	methods	method	NOUN
bracis-19021	17	24	and	and	CCONJ
bracis-19021	17	25	their	their	PRON
bracis-19021	17	26	behavior	behavior	NOUN
bracis-19021	17	27	in	in	ADP
bracis-19021	17	28	practice	practice	NOUN
bracis-19021	17	29	,	,	PUNCT
bracis-19021	17	30	often	often	ADV
bracis-19021	17	31	due	due	ADP
bracis-19021	17	32	to	to	ADP
bracis-19021	17	33	code	code	NOUN
bracis-19021	17	34	-	-	PUNCT
bracis-19021	17	35	level	level	NOUN
bracis-19021	17	36	optimizations	optimization	NOUN
bracis-19021	17	37	[	[	X
bracis-19021	17	38	6	6	NUM
bracis-19021	17	39	,	,	PUNCT
bracis-19021	17	40	11	11	NUM
bracis-19021	17	41	,	,	PUNCT
bracis-19021	17	42	13	13	NUM
bracis-19021	17	43	]	]	PUNCT
bracis-19021	17	44	.	.	PUNCT
bracis-19021	18	1	moreover	moreover	ADV
bracis-19021	18	2	,	,	PUNCT
bracis-19021	18	3	it	it	PRON
bracis-19021	18	4	is	be	AUX
bracis-19021	18	5	common	common	ADJ
bracis-19021	18	6	for	for	ADP
bracis-19021	18	7	theoretically	theoretically	ADV
bracis-19021	18	8	promising	promising	ADJ
bracis-19021	18	9	mbrl	mbrl	NOUN
bracis-19021	18	10	algorithms	algorithm	NOUN
bracis-19021	18	11	to	to	PART
bracis-19021	18	12	fail	fail	VERB
bracis-19021	18	13	in	in	ADP
bracis-19021	18	14	practice	practice	NOUN
bracis-19021	18	15	[	[	X
bracis-19021	18	16	14	14	NUM
bracis-19021	18	17	]	]	PUNCT
bracis-19021	18	18	,	,	PUNCT
bracis-19021	18	19	although	although	SCONJ
bracis-19021	18	20	negative	negative	ADJ
bracis-19021	18	21	results	result	NOUN
bracis-19021	18	22	are	be	AUX
bracis-19021	18	23	not	not	PART
bracis-19021	18	24	often	often	ADV
bracis-19021	18	25	publicized	publicize	VERB
bracis-19021	18	26	.	.	PUNCT
bracis-19021	19	1	in	in	ADP
bracis-19021	19	2	this	this	DET
bracis-19021	19	3	work	work	NOUN
bracis-19021	19	4	,	,	PUNCT
bracis-19021	19	5	we	we	PRON
bracis-19021	19	6	take	take	VERB
bracis-19021	19	7	a	a	DET
bracis-19021	19	8	step	step	NOUN
bracis-19021	19	9	towards	towards	ADP
bracis-19021	19	10	a	a	DET
bracis-19021	19	11	better	well	ADJ
bracis-19021	19	12	understanding	understanding	NOUN
bracis-19021	19	13	of	of	ADP
bracis-19021	19	14	the	the	DET
bracis-19021	19	15	core	core	NOUN
bracis-19021	19	16	tenets	tenet	NOUN
bracis-19021	19	17	behind	behind	ADP
bracis-19021	19	18	svg	svg	NOUN
bracis-19021	19	19	methods	method	NOUN
bracis-19021	19	20	by	by	ADP
bracis-19021	19	21	analyzing	analyze	VERB
bracis-19021	19	22	an	an	DET
bracis-19021	19	23	important	important	ADJ
bracis-19021	19	24	algorithm	algorithm	NOUN
bracis-19021	19	25	design	design	NOUN
bracis-19021	19	26	choice	choice	NOUN
bracis-19021	19	27	:	:	PUNCT
bracis-19021	19	28	the	the	DET
bracis-19021	19	29	gradient	gradient	ADJ
bracis-19021	19	30	estimator	estimator	NOUN
bracis-19021	19	31	computation	computation	NOUN
bracis-19021	19	32	[	[	X
bracis-19021	19	33	17	17	NUM
bracis-19021	19	34	,	,	PUNCT
bracis-19021	19	35	23	23	NUM
bracis-19021	19	36	]	]	PUNCT
bracis-19021	19	37	.	.	PUNCT
bracis-19021	20	1	we	we	PRON
bracis-19021	20	2	aim	aim	VERB
bracis-19021	20	3	to	to	PART
bracis-19021	20	4	identify	identify	VERB
bracis-19021	20	5	the	the	DET
bracis-19021	20	6	key	key	ADJ
bracis-19021	20	7	practical	practical	ADJ
bracis-19021	20	8	differences	difference	NOUN
bracis-19021	20	9	between	between	ADP
bracis-19021	20	10	the	the	DET
bracis-19021	20	11	theoretically	theoretically	ADV
bracis-19021	20	12	sound	sound	ADJ
bracis-19021	20	13	,	,	PUNCT
bracis-19021	20	14	unbiased	unbiased	ADJ
bracis-19021	20	15	formulation	formulation	NOUN
bracis-19021	20	16	from	from	ADP
bracis-19021	20	17	the	the	DET
bracis-19021	20	18	deterministic	deterministic	ADJ
bracis-19021	20	19	policy	policy	NOUN
bracis-19021	20	20	gradient	gradient	NOUN
bracis-19021	20	21	(	(	PUNCT
bracis-19021	20	22	dpg	dpg	ADJ
bracis-19021	20	23	)	)	PUNCT
bracis-19021	20	24	framework	framework	NOUN
bracis-19021	20	25	[	[	X
bracis-19021	20	26	24	24	NUM
bracis-19021	20	27	]	]	PUNCT
bracis-19021	20	28	and	and	CCONJ
bracis-19021	20	29	an	an	DET
bracis-19021	20	30	estimator	estimator	NOUN
bracis-19021	20	31	more	more	ADV
bracis-19021	20	32	often	often	ADV
bracis-19021	20	33	used	use	VERB
bracis-19021	20	34	in	in	ADP
bracis-19021	20	35	svg	svg	PROPN
bracis-19021	20	36	methods	method	NOUN
bracis-19021	20	37	[	[	X
bracis-19021	20	38	1	1	NUM
bracis-19021	20	39	,	,	PUNCT
bracis-19021	20	40	4	4	NUM
bracis-19021	20	41	,	,	PUNCT
bracis-19021	20	42	8	8	NUM
bracis-19021	20	43	]	]	PUNCT
bracis-19021	20	44	that	that	PRON
bracis-19021	20	45	deviates	deviate	VERB
bracis-19021	20	46	from	from	ADP
bracis-19021	20	47	traditional	traditional	ADJ
bracis-19021	20	48	policy	policy	NOUN
bracis-19021	20	49	gradient	gradient	NOUN
bracis-19021	20	50	theory	theory	NOUN
bracis-19021	20	51	.	.	PUNCT
bracis-19021	21	1	the	the	DET
bracis-19021	21	2	former	former	NOUN
bracis-19021	21	3	represents	represent	VERB
bracis-19021	21	4	an	an	DET
bracis-19021	21	5	approach	approach	NOUN
bracis-19021	21	6	that	that	PRON
bracis-19021	21	7	’s	’	VERB
bracis-19021	21	8	theoretically	theoretically	ADV
bracis-19021	21	9	sound	sound	ADJ
bracis-19021	21	10	,	,	PUNCT
bracis-19021	21	11	but	but	CCONJ
bracis-19021	21	12	lacks	lack	VERB
bracis-19021	21	13	empirical	empirical	ADJ
bracis-19021	21	14	validation	validation	NOUN
bracis-19021	21	15	,	,	PUNCT
bracis-19021	21	16	while	while	SCONJ
bracis-19021	21	17	the	the	DET
bracis-19021	21	18	latter	latter	NOUN
bracis-19021	21	19	generalizes	generalize	VERB
bracis-19021	21	20	the	the	DET
bracis-19021	21	21	methods	method	NOUN
bracis-19021	21	22	that	that	PRON
bracis-19021	21	23	deviate	deviate	VERB
bracis-19021	21	24	from	from	ADP
bracis-19021	21	25	the	the	DET
bracis-19021	21	26	theory	theory	NOUN
bracis-19021	21	27	,	,	PUNCT
bracis-19021	21	28	but	but	CCONJ
bracis-19021	21	29	achieve	achieve	VERB
bracis-19021	21	30	state	state	NOUN
bracis-19021	21	31	-	-	PUNCT
bracis-19021	21	32	of	of	ADP
bracis-19021	21	33	-	-	PUNCT
bracis-19021	21	34	the	the	DET
bracis-19021	21	35	-	-	PUNCT
bracis-19021	21	36	art	art	NOUN
bracis-19021	21	37	results	result	NOUN
bracis-19021	21	38	in	in	ADP
bracis-19021	21	39	continuous	continuous	ADJ
bracis-19021	21	40	control	control	NOUN
bracis-19021	21	41	benchmarks	benchmark	NOUN
bracis-19021	21	42	.	.	PUNCT
bracis-19021	22	1	we	we	PRON
bracis-19021	22	2	evaluate	evaluate	VERB
bracis-19021	22	3	gradient	gradient	ADJ
bracis-19021	22	4	estimation	estimation	NOUN
bracis-19021	22	5	quality	quality	NOUN
bracis-19021	22	6	and	and	CCONJ
bracis-19021	22	7	policy	policy	NOUN
bracis-19021	22	8	optimization	optimization	NOUN
bracis-19021	22	9	performance	performance	NOUN
bracis-19021	22	10	on	on	ADP
bracis-19021	22	11	linear	linear	ADJ
bracis-19021	22	12	quadratic	quadratic	ADJ
bracis-19021	22	13	gaussian	gaussian	NOUN
bracis-19021	22	14	(	(	PUNCT
bracis-19021	22	15	lqg	lqg	NOUN
bracis-19021	22	16	)	)	PUNCT
bracis-19021	22	17	regulator	regulator	NOUN
bracis-19021	22	18	environments	environment	NOUN
bracis-19021	22	19	.	.	PUNCT
bracis-19021	23	1	the	the	DET
bracis-19021	23	2	lqg	lqg	NOUN
bracis-19021	23	3	framework	framework	NOUN
bracis-19021	23	4	is	be	AUX
bracis-19021	23	5	extensively	extensively	ADV
bracis-19021	23	6	studied	study	VERB
bracis-19021	23	7	in	in	ADP
bracis-19021	23	8	the	the	DET
bracis-19021	23	9	optimal	optimal	ADJ
bracis-19021	23	10	control	control	NOUN
bracis-19021	23	11	literature	literature	NOUN
bracis-19021	23	12	and	and	CCONJ
bracis-19021	23	13	is	be	AUX
bracis-19021	23	14	a	a	DET
bracis-19021	23	15	special	special	ADJ
bracis-19021	23	16	class	class	NOUN
bracis-19021	23	17	of	of	ADP
bracis-19021	23	18	rl	rl	NOUN
bracis-19021	23	19	with	with	ADP
bracis-19021	23	20	continuous	continuous	ADJ
bracis-19021	23	21	actions	action	NOUN
bracis-19021	23	22	(	(	PUNCT
bracis-19021	23	23	a.k.a	a.k.a	ADJ
bracis-19021	23	24	.	.	INTJ
bracis-19021	23	25	continuous	continuous	ADJ
bracis-19021	23	26	control	control	NOUN
bracis-19021	23	27	)	)	PUNCT
bracis-19021	23	28	,	,	PUNCT
bracis-19021	23	29	where	where	SCONJ
bracis-19021	23	30	svg	svg	PROPN
bracis-19021	23	31	methods	method	NOUN
bracis-19021	23	32	seem	seem	VERB
bracis-19021	23	33	to	to	PART
bracis-19021	23	34	excel	excel	VERB
bracis-19021	23	35	.	.	PUNCT
bracis-19021	24	1	lqg	lqg	NOUN
bracis-19021	24	2	has	have	AUX
bracis-19021	24	3	been	be	AUX
bracis-19021	24	4	proposed	propose	VERB
bracis-19021	24	5	as	as	ADP
bracis-19021	24	6	a	a	DET
bracis-19021	24	7	simple	simple	ADJ
bracis-19021	24	8	,	,	PUNCT
bracis-19021	24	9	yet	yet	CCONJ
bracis-19021	24	10	nontrivial	nontrivial	ADJ
bracis-19021	24	11	class	class	NOUN
bracis-19021	24	12	of	of	ADP
bracis-19021	24	13	continuous	continuous	ADJ
bracis-19021	24	14	control	control	NOUN
bracis-19021	24	15	problems	problem	NOUN
bracis-19021	24	16	to	to	PART
bracis-19021	24	17	help	help	VERB
bracis-19021	24	18	distinguish	distinguish	VERB
bracis-19021	24	19	the	the	DET
bracis-19021	24	20	various	various	ADJ
bracis-19021	24	21	approaches	approach	NOUN
bracis-19021	24	22	to	to	PART
bracis-19021	24	23	rl	rl	VERB
bracis-19021	24	24	[	[	X
bracis-19021	24	25	21	21	NUM
bracis-19021	24	26	]	]	PUNCT
bracis-19021	24	27	.	.	PUNCT
bracis-19021	25	1	more	more	ADV
bracis-19021	25	2	importantly	importantly	ADV
bracis-19021	25	3	,	,	PUNCT
bracis-19021	25	4	lqg	lqg	PROPN
bracis-19021	25	5	has	have	VERB
bracis-19021	25	6	a	a	DET
bracis-19021	25	7	simple	simple	ADJ
bracis-19021	25	8	environment	environment	NOUN
bracis-19021	25	9	formulation	formulation	NOUN
bracis-19021	25	10	,	,	PUNCT
bracis-19021	25	11	allowing	allow	VERB
bracis-19021	25	12	us	we	PRON
bracis-19021	25	13	to	to	PART
bracis-19021	25	14	compute	compute	VERB
bracis-19021	25	15	the	the	DET
bracis-19021	25	16	ground	ground	NOUN
bracis-19021	25	17	-	-	PUNCT
bracis-19021	25	18	truth	truth	NOUN
bracis-19021	25	19	policy	policy	NOUN
bracis-19021	25	20	performance	performance	NOUN
bracis-19021	25	21	and	and	CCONJ
bracis-19021	25	22	gradient	gradient	NOUN
bracis-19021	25	23	,	,	PUNCT
bracis-19021	25	24	an	an	DET
bracis-19021	25	25	ability	ability	NOUN
bracis-19021	25	26	not	not	PART
bracis-19021	25	27	present	present	ADJ
bracis-19021	25	28	in	in	ADP
bracis-19021	25	29	more	more	ADV
bracis-19021	25	30	complicated	complicated	ADJ
bracis-19021	25	31	,	,	PUNCT
bracis-19021	25	32	nonlinear	nonlinear	ADJ
bracis-19021	25	33	benchmarks	benchmark	NOUN
bracis-19021	25	34	for	for	ADP
bracis-19021	25	35	continuous	continuous	ADJ
bracis-19021	25	36	control	control	NOUN
bracis-19021	25	37	[	[	X
bracis-19021	25	38	28	28	NUM
bracis-19021	25	39	]	]	PUNCT
bracis-19021	25	40	.	.	PUNCT
bracis-19021	26	1	thus	thus	ADV
bracis-19021	26	2	,	,	PUNCT
bracis-19021	26	3	we	we	PRON
bracis-19021	26	4	can	can	AUX
bracis-19021	26	5	perform	perform	VERB
bracis-19021	26	6	a	a	DET
bracis-19021	26	7	more	more	ADV
bracis-19021	26	8	rigorous	rigorous	ADJ
bracis-19021	26	9	empirical	empirical	ADJ
bracis-19021	26	10	assessment	assessment	NOUN
bracis-19021	26	11	of	of	ADP
bracis-19021	26	12	gradient	gradient	ADJ
bracis-19021	26	13	estimation	estimation	NOUN
bracis-19021	26	14	quality	quality	NOUN
bracis-19021	26	15	and	and	CCONJ
bracis-19021	26	16	policy	policy	NOUN
bracis-19021	26	17	optimization	optimization	NOUN
bracis-19021	26	18	performance	performance	NOUN
bracis-19021	26	19	within	within	ADP
bracis-19021	26	20	the	the	DET
bracis-19021	26	21	lqg	lqg	NOUN
bracis-19021	26	22	framework	framework	NOUN
bracis-19021	26	23	.	.	PUNCT
bracis-19021	27	1	therefore	therefore	ADV
bracis-19021	27	2	,	,	PUNCT
bracis-19021	27	3	our	our	PRON
bracis-19021	27	4	main	main	ADJ
bracis-19021	27	5	contribution	contribution	NOUN
bracis-19021	27	6	in	in	ADP
bracis-19021	27	7	this	this	DET
bracis-19021	27	8	paper	paper	NOUN
bracis-19021	27	9	is	be	AUX
bracis-19021	27	10	a	a	DET
bracis-19021	27	11	careful	careful	ADJ
bracis-19021	27	12	empirical	empirical	ADJ
bracis-19021	27	13	analysis	analysis	NOUN
bracis-19021	27	14	,	,	PUNCT
bracis-19021	27	15	using	use	VERB
bracis-19021	27	16	the	the	DET
bracis-19021	27	17	lqg	lqg	NOUN
bracis-19021	27	18	framework	framework	NOUN
bracis-19021	27	19	,	,	PUNCT
bracis-19021	27	20	of	of	ADP
bracis-19021	27	21	the	the	DET
bracis-19021	27	22	practical	practical	ADJ
bracis-19021	27	23	differences	difference	NOUN
bracis-19021	27	24	between	between	ADP
bracis-19021	27	25	gradient	gradient	ADJ
bracis-19021	27	26	estimators	estimator	NOUN
bracis-19021	27	27	for	for	ADP
bracis-19021	27	28	svg	svg	NOUN
bracis-19021	27	29	methods	method	NOUN
bracis-19021	27	30	.	.	PUNCT
bracis-19021	28	1	the	the	DET
bracis-19021	28	2	rest	rest	NOUN
bracis-19021	28	3	of	of	ADP
bracis-19021	28	4	this	this	DET
bracis-19021	28	5	paper	paper	NOUN
bracis-19021	28	6	is	be	AUX
bracis-19021	28	7	organized	organize	VERB
bracis-19021	28	8	as	as	SCONJ
bracis-19021	28	9	follows	follow	VERB
bracis-19021	28	10	.	.	PUNCT
bracis-19021	29	1	section	section	NOUN
bracis-19021	29	2	 	 	SPACE
bracis-19021	29	3	2	2	NUM
bracis-19021	29	4	outlines	outline	NOUN
bracis-19021	29	5	related	related	ADJ
bracis-19021	29	6	work	work	NOUN
bracis-19021	29	7	on	on	ADP
bracis-19021	29	8	svg	svg	NOUN
bracis-19021	29	9	methods	method	NOUN
bracis-19021	29	10	and	and	CCONJ
bracis-19021	29	11	the	the	DET
bracis-19021	29	12	gap	gap	NOUN
bracis-19021	29	13	between	between	ADP
bracis-19021	29	14	theory	theory	NOUN
bracis-19021	29	15	and	and	CCONJ
bracis-19021	29	16	practice	practice	NOUN
bracis-19021	29	17	of	of	ADP
bracis-19021	29	18	rl	rl	PROPN
bracis-19021	29	19	.	.	PROPN
bracis-19021	29	20	section	section	NOUN
bracis-19021	29	21	 	 	SPACE
bracis-19021	29	22	3	3	NUM
bracis-19021	29	23	introduces	introduce	VERB
bracis-19021	29	24	the	the	DET
bracis-19021	29	25	reader	reader	NOUN
bracis-19021	29	26	to	to	ADP
bracis-19021	29	27	the	the	DET
bracis-19021	29	28	minimal	minimal	ADJ
bracis-19021	29	29	technical	technical	ADJ
bracis-19021	29	30	background	background	NOUN
bracis-19021	29	31	on	on	ADP
bracis-19021	29	32	rl	rl	PROPN
bracis-19021	29	33	,	,	PUNCT
bracis-19021	29	34	lqg	lqg	NOUN
bracis-19021	29	35	and	and	CCONJ
bracis-19021	29	36	svg	svg	PROPN
bracis-19021	29	37	,	,	PUNCT
bracis-19021	29	38	required	require	VERB
bracis-19021	29	39	to	to	PART
bracis-19021	29	40	follow	follow	VERB
bracis-19021	29	41	our	our	PRON
bracis-19021	29	42	analysis	analysis	NOUN
bracis-19021	29	43	.	.	PUNCT
bracis-19021	30	1	section	section	NOUN
bracis-19021	30	2	 	 	SPACE
bracis-19021	30	3	4	4	NUM
bracis-19021	30	4	describes	describe	VERB
bracis-19021	30	5	the	the	DET
bracis-19021	30	6	scope	scope	NOUN
bracis-19021	30	7	of	of	ADP
bracis-19021	30	8	the	the	DET
bracis-19021	30	9	empirical	empirical	ADJ
bracis-19021	30	10	analysis	analysis	NOUN
bracis-19021	30	11	and	and	CCONJ
bracis-19021	30	12	the	the	DET
bracis-19021	30	13	experimental	experimental	ADJ
bracis-19021	30	14	setup	setup	NOUN
bracis-19021	30	15	developed	develop	VERB
bracis-19021	30	16	accordingly	accordingly	ADV
bracis-19021	30	17	.	.	PUNCT
bracis-19021	31	1	section	section	NOUN
bracis-19021	31	2	 	 	SPACE
bracis-19021	31	3	5	5	NUM
bracis-19021	31	4	contains	contain	VERB
bracis-19021	31	5	the	the	DET
bracis-19021	31	6	main	main	ADJ
bracis-19021	31	7	experiments	experiment	NOUN
bracis-19021	31	8	,	,	PUNCT
bracis-19021	31	9	results	result	NOUN
bracis-19021	31	10	,	,	PUNCT
bracis-19021	31	11	and	and	CCONJ
bracis-19021	31	12	our	our	PRON
bracis-19021	31	13	analysis	analysis	NOUN
bracis-19021	31	14	thereof	thereof	ADV
bracis-19021	31	15	.	.	PUNCT
bracis-19021	32	1	finally	finally	ADV
bracis-19021	32	2	,	,	PUNCT
bracis-19021	32	3	we	we	PRON
bracis-19021	32	4	summarize	summarize	VERB
bracis-19021	32	5	our	our	PRON
bracis-19021	32	6	observations	observation	NOUN
bracis-19021	32	7	in	in	ADP
bracis-19021	32	8	sect	sect	NOUN
bracis-19021	32	9	.	.	PUNCT
bracis-19021	32	10	 	 	SPACE
bracis-19021	32	11	6	6	NUM
bracis-19021	32	12	and	and	CCONJ
bracis-19021	32	13	point	point	VERB
bracis-19021	32	14	to	to	ADP
bracis-19021	32	15	possible	possible	ADJ
bracis-19021	32	16	future	future	ADJ
bracis-19021	32	17	work	work	NOUN
bracis-19021	32	18	to	to	PART
bracis-19021	32	19	explore	explore	VERB
bracis-19021	32	20	remaining	remain	VERB
bracis-19021	32	21	gaps	gap	NOUN
bracis-19021	32	22	in	in	ADP
bracis-19021	32	23	our	our	PRON
bracis-19021	32	24	knowledge	knowledge	NOUN
bracis-19021	32	25	of	of	ADP
bracis-19021	32	26	the	the	DET
bracis-19021	32	27	core	core	NOUN
bracis-19021	32	28	tenets	tenet	NOUN
bracis-19021	32	29	of	of	ADP
bracis-19021	32	30	svg	svg	NOUN
bracis-19021	32	31	methods	method	NOUN
bracis-19021	32	32	.	.	PUNCT
bracis-19021	33	1	2	2	NUM
bracis-19021	33	2	related	relate	VERB
bracis-19021	33	3	work	work	NOUN
bracis-19021	33	4	model	model	NOUN
bracis-19021	33	5	-	-	PUNCT
bracis-19021	33	6	based	base	VERB
bracis-19021	33	7	policy	policy	NOUN
bracis-19021	33	8	gradients	gradient	NOUN
bracis-19021	33	9	have	have	AUX
bracis-19021	33	10	been	be	AUX
bracis-19021	33	11	recently	recently	ADV
bracis-19021	33	12	used	use	VERB
bracis-19021	33	13	in	in	ADP
bracis-19021	33	14	a	a	DET
bracis-19021	33	15	variety	variety	NOUN
bracis-19021	33	16	of	of	ADP
bracis-19021	33	17	rl	rl	ADP
bracis-19021	33	18	algorithms	algorithm	NOUN
bracis-19021	33	19	for	for	ADP
bracis-19021	33	20	continuous	continuous	ADJ
bracis-19021	33	21	control	control	NOUN
bracis-19021	33	22	.	.	PUNCT
bracis-19021	34	1	the	the	DET
bracis-19021	34	2	pilco	pilco	ADJ
bracis-19021	34	3	algorithm	algorithm	NOUN
bracis-19021	34	4	is	be	AUX
bracis-19021	34	5	one	one	NUM
bracis-19021	34	6	of	of	ADP
bracis-19021	34	7	the	the	DET
bracis-19021	34	8	first	first	ADJ
bracis-19021	34	9	to	to	PART
bracis-19021	34	10	leverage	leverage	VERB
bracis-19021	34	11	a	a	DET
bracis-19021	34	12	learned	learn	VERB
bracis-19021	34	13	model	model	NOUN
bracis-19021	34	14	’s	’s	PART
bracis-19021	34	15	derivatives	derivative	NOUN
bracis-19021	34	16	to	to	PART
bracis-19021	34	17	compute	compute	VERB
bracis-19021	34	18	the	the	DET
bracis-19021	34	19	policy	policy	NOUN
bracis-19021	34	20	gradient	gradient	NOUN
bracis-19021	34	21	with	with	ADP
bracis-19021	34	22	few	few	ADJ
bracis-19021	34	23	samples	sample	NOUN
bracis-19021	34	24	,	,	PUNCT
bracis-19021	34	25	but	but	CCONJ
bracis-19021	34	26	its	its	PRON
bracis-19021	34	27	use	use	NOUN
bracis-19021	34	28	of	of	ADP
bracis-19021	34	29	gaussian	gaussian	ADJ
bracis-19021	34	30	processes	process	NOUN
bracis-19021	34	31	hinders	hinder	VERB
bracis-19021	34	32	scalability	scalability	NOUN
bracis-19021	34	33	to	to	ADP
bracis-19021	34	34	larger	large	ADJ
bracis-19021	34	35	problems	problem	NOUN
bracis-19021	34	36	[	[	X
bracis-19021	34	37	5	5	NUM
bracis-19021	34	38	]	]	PUNCT
bracis-19021	34	39	.	.	PUNCT
bracis-19021	35	1	the	the	DET
bracis-19021	35	2	original	original	ADJ
bracis-19021	35	3	svg	svg	PROPN
bracis-19021	35	4	paper	paper	NOUN
bracis-19021	35	5	introduced	introduce	VERB
bracis-19021	35	6	gradient	gradient	ADJ
bracis-19021	35	7	estimation	estimation	NOUN
bracis-19021	35	8	with	with	ADP
bracis-19021	35	9	stochastic	stochastic	ADJ
bracis-19021	35	10	neural	neural	ADJ
bracis-19021	35	11	network	network	NOUN
bracis-19021	35	12	models	model	NOUN
bracis-19021	35	13	using	use	VERB
bracis-19021	35	14	the	the	DET
bracis-19021	35	15	reparameterization	reparameterization	NOUN
bracis-19021	35	16	trick	trick	NOUN
bracis-19021	35	17	,	,	PUNCT
bracis-19021	35	18	an	an	DET
bracis-19021	35	19	approach	approach	NOUN
bracis-19021	35	20	scalable	scalable	ADJ
bracis-19021	35	21	to	to	ADP
bracis-19021	35	22	higher	higher	ADV
bracis-19021	35	23	-	-	PUNCT
bracis-19021	35	24	dimensional	dimensional	ADJ
bracis-19021	35	25	problems	problem	NOUN
bracis-19021	35	26	[	[	X
bracis-19021	35	27	9	9	NUM
bracis-19021	35	28	]	]	PUNCT
bracis-19021	35	29	.	.	PUNCT
bracis-19021	36	1	dreamer	dreamer	PROPN
bracis-19021	36	2	and	and	CCONJ
bracis-19021	36	3	imagined	imagined	ADJ
bracis-19021	36	4	value	value	NOUN
bracis-19021	36	5	gradients	gradient	NOUN
bracis-19021	36	6	explore	explore	VERB
bracis-19021	36	7	svgs	svgs	NOUN
bracis-19021	36	8	with	with	ADP
bracis-19021	36	9	latent	latent	ADJ
bracis-19021	36	10	-	-	PUNCT
bracis-19021	36	11	space	space	NOUN
bracis-19021	36	12	models	model	NOUN
bracis-19021	36	13	[	[	X
bracis-19021	36	14	2	2	NUM
bracis-19021	36	15	,	,	PUNCT
bracis-19021	36	16	8	8	NUM
bracis-19021	36	17	]	]	PUNCT
bracis-19021	36	18	.	.	PUNCT
bracis-19021	37	1	model	model	NOUN
bracis-19021	37	2	-	-	PUNCT
bracis-19021	37	3	augmented	augment	VERB
bracis-19021	37	4	actor	actor	NOUN
bracis-19021	37	5	-	-	PUNCT
bracis-19021	37	6	critic	critic	NOUN
bracis-19021	37	7	(	(	PUNCT
bracis-19021	37	8	maac	maac	NOUN
bracis-19021	37	9	)	)	PUNCT
bracis-19021	37	10	and	and	CCONJ
bracis-19021	37	11	sac	sac	NOUN
bracis-19021	37	12	-	-	PUNCT
bracis-19021	37	13	svg	svg	NOUN
bracis-19021	37	14	extend	extend	NOUN
bracis-19021	38	1	the	the	DET
bracis-19021	38	2	svg	svg	NOUN
bracis-19021	38	3	framework	framework	NOUN
bracis-19021	38	4	to	to	ADP
bracis-19021	38	5	that	that	PRON
bracis-19021	38	6	of	of	ADP
bracis-19021	38	7	maximum	maximum	ADJ
bracis-19021	38	8	-	-	PUNCT
bracis-19021	38	9	entropy	entropy	NOUN
bracis-19021	38	10	rl	rl	PRON
bracis-19021	38	11	to	to	PART
bracis-19021	38	12	incentivize	incentivize	VERB
bracis-19021	38	13	exploration	exploration	NOUN
bracis-19021	38	14	and	and	CCONJ
bracis-19021	38	15	stabilize	stabilize	VERB
bracis-19021	38	16	optimization	optimization	NOUN
bracis-19021	38	17	[	[	X
bracis-19021	38	18	1	1	NUM
bracis-19021	38	19	,	,	PUNCT
bracis-19021	38	20	4	4	NUM
bracis-19021	38	21	]	]	PUNCT
bracis-19021	38	22	.	.	PUNCT
bracis-19021	39	1	however	however	ADV
bracis-19021	39	2	,	,	PUNCT
bracis-19021	39	3	the	the	DET
bracis-19021	39	4	computation	computation	NOUN
bracis-19021	39	5	of	of	ADP
bracis-19021	39	6	the	the	DET
bracis-19021	39	7	gradient	gradient	NOUN
bracis-19021	39	8	is	be	AUX
bracis-19021	39	9	not	not	PART
bracis-19021	39	10	studied	study	VERB
bracis-19021	39	11	in	in	ADP
bracis-19021	39	12	isolation	isolation	NOUN
bracis-19021	39	13	or	or	CCONJ
bracis-19021	39	14	contrasted	contrast	VERB
bracis-19021	39	15	with	with	ADP
bracis-19021	39	16	other	other	ADJ
bracis-19021	39	17	gradient	gradient	NOUN
bracis-19021	39	18	estimation	estimation	NOUN
bracis-19021	39	19	methods	method	NOUN
bracis-19021	39	20	.	.	PUNCT
bracis-19021	40	1	our	our	PRON
bracis-19021	40	2	work	work	NOUN
bracis-19021	40	3	analyses	analyse	VERB
bracis-19021	40	4	,	,	PUNCT
bracis-19021	40	5	in	in	ADP
bracis-19021	40	6	detail	detail	NOUN
bracis-19021	40	7	,	,	PUNCT
bracis-19021	40	8	the	the	DET
bracis-19021	40	9	gradient	gradient	ADJ
bracis-19021	40	10	computation	computation	NOUN
bracis-19021	40	11	used	use	VERB
bracis-19021	40	12	in	in	ADP
bracis-19021	40	13	recent	recent	ADJ
bracis-19021	40	14	svg	svg	NOUN
bracis-19021	40	15	-	-	PUNCT
bracis-19021	40	16	style	style	NOUN
bracis-19021	40	17	algorithms	algorithm	NOUN
bracis-19021	40	18	that	that	PRON
bracis-19021	40	19	show	show	VERB
bracis-19021	40	20	a	a	DET
bracis-19021	40	21	high	high	ADJ
bracis-19021	40	22	level	level	NOUN
bracis-19021	40	23	of	of	ADP
bracis-19021	40	24	performance	performance	NOUN
bracis-19021	40	25	with	with	ADP
bracis-19021	40	26	efficient	efficient	ADJ
bracis-19021	40	27	use	use	NOUN
bracis-19021	40	28	of	of	ADP
bracis-19021	40	29	data	datum	NOUN
bracis-19021	40	30	[	[	X
bracis-19021	40	31	1	1	NUM
bracis-19021	40	32	,	,	PUNCT
bracis-19021	40	33	4	4	NUM
bracis-19021	40	34	,	,	PUNCT
bracis-19021	40	35	8	8	NUM
bracis-19021	40	36	]	]	PUNCT
bracis-19021	40	37	.	.	PUNCT
bracis-19021	41	1	a	a	DET
bracis-19021	41	2	few	few	ADJ
bracis-19021	41	3	recent	recent	ADJ
bracis-19021	41	4	works	work	NOUN
bracis-19021	41	5	have	have	AUX
bracis-19021	41	6	analyzed	analyze	VERB
bracis-19021	41	7	the	the	DET
bracis-19021	41	8	gap	gap	NOUN
bracis-19021	41	9	between	between	ADP
bracis-19021	41	10	theory	theory	NOUN
bracis-19021	41	11	and	and	CCONJ
bracis-19021	41	12	practice	practice	NOUN
bracis-19021	41	13	in	in	ADP
bracis-19021	41	14	different	different	ADJ
bracis-19021	41	15	areas	area	NOUN
bracis-19021	41	16	of	of	ADP
bracis-19021	41	17	rl	rl	PROPN
bracis-19021	41	18	,	,	PUNCT
bracis-19021	41	19	highlighting	highlight	VERB
bracis-19021	41	20	our	our	PRON
bracis-19021	41	21	poor	poor	ADJ
bracis-19021	41	22	understanding	understanding	NOUN
bracis-19021	41	23	of	of	ADP
bracis-19021	41	24	current	current	ADJ
bracis-19021	41	25	methods	method	NOUN
bracis-19021	41	26	.	.	PUNCT
bracis-19021	42	1	reliability	reliability	NOUN
bracis-19021	42	2	and	and	CCONJ
bracis-19021	42	3	reproducibility	reproducibility	NOUN
bracis-19021	42	4	concerns	concern	NOUN
bracis-19021	42	5	regarding	regard	VERB
bracis-19021	42	6	modern	modern	ADJ
bracis-19021	42	7	rl	rl	NOUN
bracis-19021	42	8	methods	method	NOUN
bracis-19021	42	9	have	have	AUX
bracis-19021	42	10	been	be	AUX
bracis-19021	42	11	raised	raise	VERB
bracis-19021	42	12	by	by	ADP
bracis-19021	42	13	several	several	ADJ
bracis-19021	42	14	works	work	NOUN
bracis-19021	42	15	[	[	X
bracis-19021	42	16	3	3	NUM
bracis-19021	42	17	,	,	PUNCT
bracis-19021	42	18	10	10	NUM
bracis-19021	42	19	,	,	PUNCT
bracis-19021	42	20	12	12	NUM
bracis-19021	42	21	]	]	PUNCT
bracis-19021	42	22	,	,	PUNCT
bracis-19021	42	23	indicating	indicate	VERB
bracis-19021	42	24	a	a	DET
bracis-19021	42	25	disconnect	disconnect	NOUN
bracis-19021	42	26	between	between	ADP
bracis-19021	42	27	the	the	DET
bracis-19021	42	28	theory	theory	NOUN
bracis-19021	42	29	motivating	motivate	VERB
bracis-19021	42	30	these	these	DET
bracis-19021	42	31	algorithms	algorithm	NOUN
bracis-19021	42	32	and	and	CCONJ
bracis-19021	42	33	their	their	PRON
bracis-19021	42	34	behavior	behavior	NOUN
bracis-19021	42	35	in	in	ADP
bracis-19021	42	36	practice	practice	NOUN
bracis-19021	42	37	.	.	PUNCT
bracis-19021	43	1	code	code	NOUN
bracis-19021	43	2	-	-	PUNCT
bracis-19021	43	3	level	level	NOUN
bracis-19021	43	4	optimizations	optimization	NOUN
bracis-19021	43	5	have	have	AUX
bracis-19021	43	6	been	be	AUX
bracis-19021	43	7	found	find	VERB
bracis-19021	43	8	to	to	PART
bracis-19021	43	9	contribute	contribute	VERB
bracis-19021	43	10	more	more	ADV
bracis-19021	43	11	to	to	ADP
bracis-19021	43	12	successful	successful	ADJ
bracis-19021	43	13	policy	policy	NOUN
bracis-19021	43	14	gradient	gradient	NOUN
bracis-19021	43	15	methods	method	NOUN
bracis-19021	43	16	than	than	ADP
bracis-19021	43	17	the	the	DET
bracis-19021	43	18	choice	choice	NOUN
bracis-19021	43	19	of	of	ADP
bracis-19021	43	20	general	general	ADJ
bracis-19021	43	21	training	training	NOUN
bracis-19021	43	22	algorithm	algorithm	NOUN
bracis-19021	43	23	[	[	X
bracis-19021	43	24	6	6	NUM
bracis-19021	43	25	]	]	PUNCT
bracis-19021	43	26	.	.	PUNCT
bracis-19021	44	1	closest	close	ADJ
bracis-19021	44	2	to	to	ADP
bracis-19021	44	3	our	our	PRON
bracis-19021	44	4	work	work	NOUN
bracis-19021	44	5	is	be	AUX
bracis-19021	44	6	that	that	PRON
bracis-19021	44	7	of	of	ADP
bracis-19021	44	8	ilyas	ilyas	PROPN
bracis-19021	44	9	et	et	PROPN
bracis-19021	44	10	al	al	PROPN
bracis-19021	44	11	.	.	PUNCT
bracis-19021	45	1	[	[	X
bracis-19021	45	2	11	11	NUM
bracis-19021	45	3	]	]	PUNCT
bracis-19021	45	4	,	,	PUNCT
bracis-19021	45	5	which	which	PRON
bracis-19021	45	6	shows	show	VERB
bracis-19021	45	7	that	that	SCONJ
bracis-19021	45	8	model	model	ADJ
bracis-19021	45	9	-	-	PUNCT
bracis-19021	45	10	free	free	ADJ
bracis-19021	45	11	policy	policy	NOUN
bracis-19021	45	12	gradient	gradient	NOUN
bracis-19021	45	13	algorithms	algorithm	NOUN
bracis-19021	45	14	succeed	succeed	VERB
bracis-19021	45	15	in	in	ADP
bracis-19021	45	16	optimizing	optimize	VERB
bracis-19021	45	17	the	the	DET
bracis-19021	45	18	policy	policy	NOUN
bracis-19021	45	19	despite	despite	SCONJ
bracis-19021	45	20	having	have	VERB
bracis-19021	45	21	poor	poor	ADJ
bracis-19021	45	22	gradient	gradient	ADJ
bracis-19021	45	23	estimation	estimation	NOUN
bracis-19021	45	24	quality	quality	NOUN
bracis-19021	45	25	metrics	metric	NOUN
bracis-19021	45	26	in	in	ADP
bracis-19021	45	27	the	the	DET
bracis-19021	45	28	relevant	relevant	ADJ
bracis-19021	45	29	sample	sample	NOUN
bracis-19021	45	30	regime	regime	NOUN
bracis-19021	45	31	.	.	PUNCT
bracis-19021	46	1	our	our	PRON
bracis-19021	46	2	work	work	NOUN
bracis-19021	46	3	is	be	AUX
bracis-19021	46	4	,	,	PUNCT
bracis-19021	46	5	to	to	ADP
bracis-19021	46	6	the	the	DET
bracis-19021	46	7	best	good	ADJ
bracis-19021	46	8	of	of	ADP
bracis-19021	46	9	our	our	PRON
bracis-19021	46	10	knowledge	knowledge	NOUN
bracis-19021	46	11	,	,	PUNCT
bracis-19021	46	12	the	the	DET
bracis-19021	46	13	first	first	ADJ
bracis-19021	46	14	to	to	PART
bracis-19021	46	15	propose	propose	VERB
bracis-19021	46	16	a	a	DET
bracis-19021	46	17	fine	fine	ADV
bracis-19021	46	18	-	-	PUNCT
bracis-19021	46	19	grained	grain	VERB
bracis-19021	46	20	analysis	analysis	NOUN
bracis-19021	46	21	of	of	ADP
bracis-19021	46	22	gradient	gradient	ADJ
bracis-19021	46	23	estimation	estimation	NOUN
bracis-19021	46	24	in	in	ADP
bracis-19021	46	25	svg	svg	PROPN
bracis-19021	46	26	methods	method	NOUN
bracis-19021	46	27	using	use	VERB
bracis-19021	46	28	lqgs	lqg	NOUN
bracis-19021	46	29	to	to	PART
bracis-19021	46	30	provide	provide	VERB
bracis-19021	46	31	solid	solid	ADJ
bracis-19021	46	32	references	reference	NOUN
bracis-19021	46	33	of	of	ADP
bracis-19021	46	34	their	their	PRON
bracis-19021	46	35	expected	expect	VERB
bracis-19021	46	36	behavior	behavior	NOUN
bracis-19021	46	37	.	.	PUNCT
bracis-19021	47	1	3	3	NUM
bracis-19021	47	2	background	background	NOUN
bracis-19021	47	3	3.1	3.1	NUM
bracis-19021	47	4	a	a	DET
bracis-19021	47	5	brief	brief	ADJ
bracis-19021	47	6	introduction	introduction	NOUN
bracis-19021	47	7	to	to	PART
bracis-19021	47	8	rl	rl	VERB
bracis-19021	47	9	we	we	PRON
bracis-19021	47	10	consider	consider	VERB
bracis-19021	47	11	the	the	DET
bracis-19021	47	12	agent	agent	NOUN
bracis-19021	47	13	-	-	PUNCT
bracis-19021	47	14	environment	environment	NOUN
bracis-19021	47	15	interaction	interaction	NOUN
bracis-19021	47	16	modeled	model	VERB
bracis-19021	47	17	as	as	ADP
bracis-19021	47	18	a	a	DET
bracis-19021	47	19	continuous	continuous	ADJ
bracis-19021	47	20	markov	markov	NOUN
bracis-19021	47	21	decision	decision	NOUN
bracis-19021	47	22	process	process	NOUN
bracis-19021	47	23	(	(	PUNCT
bracis-19021	47	24	mdp	mdp	NOUN
bracis-19021	47	25	)	)	PUNCT
bracis-19021	48	1	[	[	X
bracis-19021	48	2	25	25	NUM
bracis-19021	48	3	]	]	PUNCT
bracis-19021	48	4	,	,	PUNCT
bracis-19021	48	5	defined	define	VERB
bracis-19021	48	6	as	as	ADP
bracis-19021	48	7	the	the	DET
bracis-19021	48	8	tuple	tuple	PROPN
bracis-19021	48	9	\(({\mathcal	\(({\mathcal	PROPN
bracis-19021	48	10	s	s	NOUN
bracis-19021	48	11	}	}	PUNCT
bracis-19021	48	12	,	,	PUNCT
bracis-19021	48	13	{	{	PUNCT
bracis-19021	48	14	\mathcal	\mathcal	NOUN
bracis-19021	48	15	a	a	X
bracis-19021	48	16	}	}	PUNCT
bracis-19021	48	17	,	,	PUNCT
bracis-19021	48	18	r^	r^	NOUN
bracis-19021	48	19	{	{	PUNCT
bracis-19021	48	20	}	}	PUNCT
bracis-19021	48	21	,	,	PUNCT
bracis-19021	48	22	p^	p^	PROPN
bracis-19021	48	23	{	{	PUNCT
bracis-19021	48	24	*	*	PUNCT
bracis-19021	48	25	}	}	PUNCT
bracis-19021	48	26	,	,	PUNCT
bracis-19021	48	27	{	{	PUNCT
bracis-19021	48	28	\rho	\rho	NOUN
bracis-19021	48	29	}	}	PUNCT
bracis-19021	48	30	,	,	PUNCT
bracis-19021	48	31	h)\	h)\	PROPN
bracis-19021	48	32	)	)	PUNCT
bracis-19021	48	33	,	,	PUNCT
bracis-19021	48	34	with	with	ADP
bracis-19021	48	35	each	each	DET
bracis-19021	48	36	component	component	NOUN
bracis-19021	48	37	described	describe	VERB
bracis-19021	48	38	in	in	ADP
bracis-19021	48	39	what	what	PRON
bracis-19021	48	40	follows	follow	VERB
bracis-19021	48	41	.	.	PUNCT
bracis-19021	49	1	interaction	interaction	NOUN
bracis-19021	49	2	with	with	ADP
bracis-19021	49	3	the	the	DET
bracis-19021	49	4	environment	environment	NOUN
bracis-19021	49	5	occurs	occur	VERB
bracis-19021	49	6	in	in	ADP
bracis-19021	49	7	a	a	DET
bracis-19021	49	8	sequence	sequence	NOUN
bracis-19021	49	9	of	of	ADP
bracis-19021	49	10	discrete	discrete	ADJ
bracis-19021	49	11	timesteps	timestep	NOUN
bracis-19021	49	12	,	,	PUNCT
bracis-19021	49	13	where	where	SCONJ
bracis-19021	49	14	\(h\in	\(h\in	X
bracis-19021	49	15	{	{	PUNCT
bracis-19021	49	16	\mathbb	\mathbb	PROPN
bracis-19021	49	17	n}\	n}\	NUM
bracis-19021	49	18	)	)	PUNCT
bracis-19021	49	19	denotes	denote	VERB
bracis-19021	49	20	the	the	DET
bracis-19021	49	21	time	time	NOUN
bracis-19021	49	22	horizon	horizon	NOUN
bracis-19021	49	23	,	,	PUNCT
bracis-19021	49	24	after	after	ADP
bracis-19021	49	25	which	which	PRON
bracis-19021	49	26	an	an	DET
bracis-19021	49	27	episode	episode	NOUN
bracis-19021	49	28	of	of	ADP
bracis-19021	49	29	interaction	interaction	NOUN
bracis-19021	49	30	is	be	AUX
bracis-19021	49	31	over	over	ADV
bracis-19021	49	32	.	.	PUNCT
bracis-19021	50	1	at	at	ADP
bracis-19021	50	2	every	every	DET
bracis-19021	50	3	timestep	timestep	NOUN
bracis-19021	50	4	t	t	PROPN
bracis-19021	50	5	,	,	PUNCT
bracis-19021	50	6	the	the	DET
bracis-19021	50	7	agent	agent	NOUN
bracis-19021	50	8	observes	observe	VERB
bracis-19021	50	9	the	the	DET
bracis-19021	50	10	current	current	ADJ
bracis-19021	50	11	state	state	NOUN
bracis-19021	50	12	\({\mathbf	\({\mathbf	NOUN
bracis-19021	50	13	{	{	PUNCT
bracis-19021	50	14	s}}_t\in	s}}_t\in	X
bracis-19021	50	15	{	{	PUNCT
bracis-19021	50	16	\mathcal	\mathcal	PROPN
bracis-19021	50	17	s}\subseteq	s}\subseteq	PROPN
bracis-19021	50	18	{	{	PUNCT
bracis-19021	50	19	\mathbb	\mathbb	PROPN
bracis-19021	50	20	r}^{n}\	r}^{n}\	PROPN
bracis-19021	50	21	)	)	PUNCT
bracis-19021	50	22	from	from	ADP
bracis-19021	50	23	the	the	DET
bracis-19021	50	24	set	set	NOUN
bracis-19021	50	25	of	of	ADP
bracis-19021	50	26	possible	possible	ADJ
bracis-19021	50	27	states	state	NOUN
bracis-19021	50	28	of	of	ADP
bracis-19021	50	29	the	the	DET
bracis-19021	50	30	environment	environment	NOUN
bracis-19021	50	31	.	.	PUNCT
bracis-19021	51	1	it	it	PRON
bracis-19021	51	2	must	must	AUX
bracis-19021	51	3	then	then	ADV
bracis-19021	51	4	select	select	VERB
bracis-19021	51	5	an	an	DET
bracis-19021	51	6	action	action	NOUN
bracis-19021	51	7	\({\mathbf	\({\mathbf	NOUN
bracis-19021	51	8	{	{	PUNCT
bracis-19021	51	9	a}}_t\in	a}}_t\in	X
bracis-19021	51	10	{	{	PUNCT
bracis-19021	51	11	\mathcal	\mathcal	PROPN
bracis-19021	51	12	a}\subseteq	a}\subseteq	PROPN
bracis-19021	51	13	{	{	PUNCT
bracis-19021	51	14	\mathbb	\mathbb	PROPN
bracis-19021	51	15	r}^{d}\	r}^{d}\	PROPN
bracis-19021	51	16	)	)	PUNCT
bracis-19021	51	17	from	from	ADP
bracis-19021	51	18	the	the	DET
bracis-19021	51	19	set	set	NOUN
bracis-19021	51	20	of	of	ADP
bracis-19021	51	21	possible	possible	ADJ
bracis-19021	51	22	actions	action	NOUN
bracis-19021	51	23	to	to	PART
bracis-19021	51	24	be	be	AUX
bracis-19021	51	25	executed	execute	VERB
bracis-19021	51	26	.	.	PUNCT
bracis-19021	52	1	the	the	DET
bracis-19021	52	2	environment	environment	NOUN
bracis-19021	52	3	then	then	ADV
bracis-19021	52	4	transitions	transition	NOUN
bracis-19021	52	5	to	to	ADP
bracis-19021	52	6	the	the	DET
bracis-19021	52	7	next	next	ADJ
bracis-19021	52	8	state	state	NOUN
bracis-19021	52	9	by	by	ADP
bracis-19021	52	10	sampling	sample	VERB
bracis-19021	52	11	from	from	ADP
bracis-19021	52	12	the	the	DET
bracis-19021	52	13	transition	transition	NOUN
bracis-19021	52	14	probability	probability	NOUN
bracis-19021	52	15	kernel	kernel	NOUN
bracis-19021	52	16	,	,	PUNCT
bracis-19021	52	17	,	,	PUNCT
bracis-19021	52	18	and	and	CCONJ
bracis-19021	52	19	emits	emit	VERB
bracis-19021	52	20	a	a	DET
bracis-19021	52	21	reward	reward	NOUN
bracis-19021	52	22	signal	signal	NOUN
bracis-19021	52	23	using	use	VERB
bracis-19021	52	24	its	its	PRON
bracis-19021	52	25	reward	reward	NOUN
bracis-19021	52	26	function	function	NOUN
bracis-19021	52	27	,	,	PUNCT
bracis-19021	52	28	\(r_{t+1	\(r_{t+1	PROPN
bracis-19021	52	29	}	}	PUNCT
bracis-19021	52	30	=	=	PUNCT
bracis-19021	52	31	r^{}({\mathbf	r^{}({\mathbf	NOUN
bracis-19021	52	32	{	{	PUNCT
bracis-19021	52	33	s}}_t	s}}_t	NOUN
bracis-19021	52	34	,	,	PUNCT
bracis-19021	52	35	{	{	PUNCT
bracis-19021	52	36	\mathbf	\mathbf	NOUN
bracis-19021	52	37	{	{	PUNCT
bracis-19021	52	38	a}}_t)\	a}}_t)\	NUM
bracis-19021	52	39	)	)	PUNCT
bracis-19021	52	40	.	.	PUNCT
bracis-19021	53	1	the	the	DET
bracis-19021	53	2	initial	initial	ADJ
bracis-19021	53	3	state	state	NOUN
bracis-19021	53	4	is	be	AUX
bracis-19021	53	5	sampled	sample	VERB
bracis-19021	53	6	from	from	ADP
bracis-19021	53	7	the	the	DET
bracis-19021	53	8	initial	initial	ADJ
bracis-19021	53	9	state	state	NOUN
bracis-19021	53	10	distribution	distribution	NOUN
bracis-19021	53	11	,	,	PUNCT
bracis-19021	53	12	\({\mathbf	\({\mathbf	NOUN
bracis-19021	53	13	{	{	PUNCT
bracis-19021	53	14	s}}_0	s}}_0	NOUN
bracis-19021	53	15	\sim	\sim	NOUN
bracis-19021	53	16	{	{	PUNCT
bracis-19021	53	17	\rho	\rho	PROPN
bracis-19021	53	18	}	}	PUNCT
bracis-19021	53	19	(	(	PUNCT
bracis-19021	53	20	\cdot	\cdot	NOUN
bracis-19021	53	21	)	)	PUNCT
bracis-19021	53	22	\	\	PROPN
bracis-19021	53	23	)	)	PUNCT
bracis-19021	53	24	.	.	PUNCT
bracis-19021	54	1	a	a	DET
bracis-19021	54	2	policy	policy	NOUN
bracis-19021	54	3	defines	define	VERB
bracis-19021	54	4	a	a	DET
bracis-19021	54	5	mapping	mapping	NOUN
bracis-19021	54	6	from	from	ADP
bracis-19021	54	7	environment	environment	NOUN
bracis-19021	54	8	states	state	NOUN
bracis-19021	54	9	to	to	ADP
bracis-19021	54	10	actions	action	NOUN
bracis-19021	54	11	,	,	PUNCT
bracis-19021	54	12	\({\mathbf	\({\mathbf	NOUN
bracis-19021	54	13	{	{	PUNCT
bracis-19021	54	14	a}}_t	a}}_t	NOUN
bracis-19021	54	15	=	=	X
bracis-19021	54	16	{	{	PUNCT
bracis-19021	54	17	\mu	\mu	PROPN
bracis-19021	54	18	}	}	PUNCT
bracis-19021	54	19	(	(	PUNCT
bracis-19021	54	20	{	{	PUNCT
bracis-19021	54	21	\mathbf	\mathbf	PROPN
bracis-19021	54	22	{	{	PUNCT
bracis-19021	54	23	s}}_t)\	s}}_t)\	NOUN
bracis-19021	54	24	)	)	PUNCT
bracis-19021	54	25	.	.	PUNCT
bracis-19021	55	1	the	the	DET
bracis-19021	55	2	objective	objective	NOUN
bracis-19021	55	3	of	of	ADP
bracis-19021	55	4	an	an	DET
bracis-19021	55	5	rl	rl	NOUN
bracis-19021	55	6	agent	agent	NOUN
bracis-19021	55	7	is	be	AUX
bracis-19021	55	8	to	to	PART
bracis-19021	55	9	find	find	VERB
bracis-19021	55	10	a	a	DET
bracis-19021	55	11	policy	policy	NOUN
bracis-19021	55	12	that	that	PRON
bracis-19021	55	13	produces	produce	VERB
bracis-19021	55	14	the	the	DET
bracis-19021	55	15	highest	high	ADJ
bracis-19021	55	16	cumulative	cumulative	ADJ
bracis-19021	55	17	reward	reward	NOUN
bracis-19021	55	18	,	,	PUNCT
bracis-19021	55	19	or	or	CCONJ
bracis-19021	55	20	return	return	VERB
bracis-19021	55	21	,	,	PUNCT
bracis-19021	55	22	from	from	ADP
bracis-19021	55	23	the	the	DET
bracis-19021	55	24	initial	initial	ADJ
bracis-19021	55	25	state	state	NOUN
bracis-19021	55	26	:	:	PUNCT
bracis-19021	55	27	.	.	PUNCT
bracis-19021	56	1	here	here	ADV
bracis-19021	56	2	,	,	PUNCT
bracis-19021	56	3	the	the	DET
bracis-19021	56	4	expectation	expectation	NOUN
bracis-19021	56	5	is	be	AUX
bracis-19021	56	6	implicitly	implicitly	ADV
bracis-19021	56	7	w.r.t	w.r.t	VERB
bracis-19021	56	8	.	.	PUNCT
bracis-19021	57	1	the	the	DET
bracis-19021	57	2	initial	initial	ADJ
bracis-19021	57	3	state	state	NOUN
bracis-19021	57	4	distribution	distribution	NOUN
bracis-19021	57	5	and	and	CCONJ
bracis-19021	57	6	the	the	DET
bracis-19021	57	7	sequential	sequential	ADJ
bracis-19021	57	8	application	application	NOUN
bracis-19021	57	9	of	of	ADP
bracis-19021	57	10	.	.	PUNCT
bracis-19021	58	1	the	the	DET
bracis-19021	58	2	key	key	ADJ
bracis-19021	58	3	difference	difference	NOUN
bracis-19021	58	4	between	between	ADP
bracis-19021	58	5	optimal	optimal	ADJ
bracis-19021	58	6	control	control	NOUN
bracis-19021	58	7	and	and	CCONJ
bracis-19021	58	8	rl	rl	PRON
bracis-19021	58	9	,	,	PUNCT
bracis-19021	58	10	both	both	DET
bracis-19021	58	11	frameworks	framework	NOUN
bracis-19021	58	12	for	for	ADP
bracis-19021	58	13	optimal	optimal	ADJ
bracis-19021	58	14	sequential	sequential	ADJ
bracis-19021	58	15	decision	decision	NOUN
bracis-19021	58	16	making	making	NOUN
bracis-19021	58	17	,	,	PUNCT
bracis-19021	58	18	is	be	AUX
bracis-19021	58	19	that	that	SCONJ
bracis-19021	58	20	in	in	ADP
bracis-19021	58	21	the	the	DET
bracis-19021	58	22	former	former	ADJ
bracis-19021	58	23	the	the	DET
bracis-19021	58	24	agent	agent	NOUN
bracis-19021	58	25	has	have	VERB
bracis-19021	58	26	access	access	NOUN
bracis-19021	58	27	to	to	ADP
bracis-19021	58	28	the	the	DET
bracis-19021	58	29	full	full	ADJ
bracis-19021	58	30	mdp	mdp	NOUN
bracis-19021	58	31	,	,	PUNCT
bracis-19021	58	32	while	while	SCONJ
bracis-19021	58	33	in	in	ADP
bracis-19021	58	34	the	the	DET
bracis-19021	58	35	latter	latter	ADJ
bracis-19021	58	36	the	the	DET
bracis-19021	58	37	agent	agent	NOUN
bracis-19021	58	38	only	only	ADV
bracis-19021	58	39	knows	know	VERB
bracis-19021	58	40	the	the	DET
bracis-19021	58	41	state	state	NOUN
bracis-19021	58	42	and	and	CCONJ
bracis-19021	58	43	action	action	NOUN
bracis-19021	58	44	space	space	NOUN
bracis-19021	58	45	and	and	CCONJ
bracis-19021	58	46	has	have	VERB
bracis-19021	58	47	to	to	PART
bracis-19021	58	48	learn	learn	VERB
bracis-19021	58	49	its	its	PRON
bracis-19021	58	50	policy	policy	NOUN
bracis-19021	58	51	by	by	ADP
bracis-19021	58	52	trial	trial	NOUN
bracis-19021	58	53	-	-	PUNCT
bracis-19021	58	54	and	and	CCONJ
bracis-19021	58	55	-	-	PUNCT
bracis-19021	58	56	error	error	NOUN
bracis-19021	58	57	in	in	ADP
bracis-19021	58	58	the	the	DET
bracis-19021	58	59	environment	environment	NOUN
bracis-19021	58	60	.	.	PUNCT
bracis-19021	59	1	3.2	3.2	NUM
bracis-19021	59	2	linear	linear	ADJ
bracis-19021	59	3	quadratic	quadratic	ADJ
bracis-19021	59	4	gaussian	gaussian	ADJ
bracis-19021	59	5	regulator	regulator	NOUN
bracis-19021	59	6	the	the	DET
bracis-19021	59	7	lqg	lqg	NOUN
bracis-19021	59	8	is	be	AUX
bracis-19021	59	9	a	a	DET
bracis-19021	59	10	special	special	ADJ
bracis-19021	59	11	class	class	NOUN
bracis-19021	59	12	of	of	ADP
bracis-19021	59	13	continuous	continuous	ADJ
bracis-19021	59	14	mdp	mdp	NOUN
bracis-19021	59	15	in	in	ADP
bracis-19021	59	16	which	which	PRON
bracis-19021	59	17	the	the	DET
bracis-19021	59	18	transition	transition	NOUN
bracis-19021	59	19	kernel	kernel	NOUN
bracis-19021	59	20	is	be	AUX
bracis-19021	59	21	linear	linear	ADJ
bracis-19021	59	22	gaussian	gaussian	NOUN
bracis-19021	59	23	and	and	CCONJ
bracis-19021	59	24	the	the	DET
bracis-19021	59	25	reward	reward	NOUN
bracis-19021	59	26	function	function	NOUN
bracis-19021	59	27	is	be	AUX
bracis-19021	59	28	quadratic	quadratic	ADJ
bracis-19021	59	29	concave	concave	NOUN
bracis-19021	60	1	[	[	X
bracis-19021	60	2	27	27	NUM
bracis-19021	60	3	]	]	SYM
bracis-19021	60	4	:	:	PUNCT
bracis-19021	60	5	(	(	PUNCT
bracis-19021	60	6	1	1	X
bracis-19021	60	7	)	)	PUNCT
bracis-19021	60	8	(	(	PUNCT
bracis-19021	60	9	2	2	X
bracis-19021	60	10	)	)	PUNCT
bracis-19021	60	11	lqgs	lqg	NOUN
bracis-19021	60	12	are	be	AUX
bracis-19021	60	13	often	often	ADV
bracis-19021	60	14	used	use	VERB
bracis-19021	60	15	as	as	ADP
bracis-19021	60	16	a	a	DET
bracis-19021	60	17	discretization	discretization	NOUN
bracis-19021	60	18	of	of	ADP
bracis-19021	60	19	continuous	continuous	ADJ
bracis-19021	60	20	-	-	PUNCT
bracis-19021	60	21	time	time	NOUN
bracis-19021	60	22	dynamics	dynamic	NOUN
bracis-19021	60	23	described	describe	VERB
bracis-19021	60	24	as	as	ADP
bracis-19021	60	25	linear	linear	PROPN
bracis-19021	60	26	differential	differential	ADJ
bracis-19021	60	27	equations	equation	NOUN
bracis-19021	60	28	,	,	PUNCT
bracis-19021	60	29	such	such	ADJ
bracis-19021	60	30	as	as	ADP
bracis-19021	60	31	those	those	PRON
bracis-19021	60	32	of	of	ADP
bracis-19021	60	33	physical	physical	ADJ
bracis-19021	60	34	systems	system	NOUN
bracis-19021	60	35	.	.	PUNCT
bracis-19021	61	1	an	an	DET
bracis-19021	61	2	important	important	ADJ
bracis-19021	61	3	class	class	NOUN
bracis-19021	61	4	of	of	ADP
bracis-19021	61	5	policies	policy	NOUN
bracis-19021	61	6	in	in	ADP
bracis-19021	61	7	this	this	DET
bracis-19021	61	8	context	context	NOUN
bracis-19021	61	9	is	be	AUX
bracis-19021	61	10	that	that	PRON
bracis-19021	61	11	of	of	ADP
bracis-19021	61	12	time	time	NOUN
bracis-19021	61	13	-	-	PUNCT
bracis-19021	61	14	varying	vary	VERB
bracis-19021	61	15	(	(	PUNCT
bracis-19021	61	16	a.k.a	a.k.a	ADJ
bracis-19021	61	17	.	.	PUNCT
bracis-19021	61	18	nonstationary	nonstationary	ADJ
bracis-19021	61	19	)	)	PUNCT
bracis-19021	61	20	linear	linear	ADJ
bracis-19021	61	21	policies	policy	NOUN
bracis-19021	61	22	:	:	PUNCT
bracis-19021	61	23	$	$	SYM
bracis-19021	61	24	$	$	SYM
bracis-19021	61	25	\begin{aligned	\begin{aligne	VERB
bracis-19021	61	26	}	}	PUNCT
bracis-19021	61	27	{	{	PUNCT
bracis-19021	61	28	\mu	\mu	PROPN
bracis-19021	61	29	}	}	PUNCT
bracis-19021	61	30	_	_	PUNCT
bracis-19021	61	31	{	{	PUNCT
bracis-19021	61	32	{	{	PUNCT
bracis-19021	61	33	\theta	\theta	PROPN
bracis-19021	61	34	}	}	PUNCT
bracis-19021	61	35	}	}	PUNCT
bracis-19021	61	36	(	(	PUNCT
bracis-19021	61	37	{	{	PUNCT
bracis-19021	61	38	\mathbf	\mathbf	PROPN
bracis-19021	61	39	{	{	PUNCT
bracis-19021	61	40	s}}_t	s}}_t	NOUN
bracis-19021	61	41	)	)	PUNCT
bracis-19021	62	1	=	=	SYM
bracis-19021	62	2	\mathbf	\mathbf	PROPN
bracis-19021	62	3	{	{	PUNCT
bracis-19021	62	4	k}_t{\mathbf	k}_t{\mathbf	NOUN
bracis-19021	62	5	{	{	PUNCT
bracis-19021	62	6	s}}_t	s}}_t	NOUN
bracis-19021	62	7	+	+	CCONJ
bracis-19021	62	8	\mathbf	\mathbf	PROPN
bracis-19021	62	9	{	{	PUNCT
bracis-19021	62	10	k}_t\	k}_t\	NOUN
bracis-19021	62	11	,	,	PUNCT
bracis-19021	62	12	,	,	PUNCT
bracis-19021	62	13	\end{aligned}$$	\end{aligned}$$	X
bracis-19021	62	14	(	(	PUNCT
bracis-19021	62	15	3	3	NUM
bracis-19021	62	16	)	)	PUNCT
bracis-19021	62	17	where	where	SCONJ
bracis-19021	62	18	\(\mathbf	\(\mathbf	VERB
bracis-19021	62	19	{	{	PUNCT
bracis-19021	62	20	k}_t	k}_t	PROPN
bracis-19021	62	21	\in	\in	PROPN
bracis-19021	62	22	{	{	PUNCT
bracis-19021	62	23	\mathbb	\mathbb	PROPN
bracis-19021	62	24	r}^{d\times	r}^{d\time	NOUN
bracis-19021	62	25	n}\	n}\	NUM
bracis-19021	62	26	)	)	PUNCT
bracis-19021	62	27	and	and	CCONJ
bracis-19021	62	28	\(\mathbf	\(\mathbf	VERB
bracis-19021	62	29	{	{	PUNCT
bracis-19021	62	30	k}_t	k}_t	PROPN
bracis-19021	62	31	\in	\in	PROPN
bracis-19021	62	32	{	{	PUNCT
bracis-19021	62	33	\mathbb	\mathbb	PROPN
bracis-19021	62	34	r}^{d}\	r}^{d}\	NOUN
bracis-19021	62	35	)	)	PUNCT
bracis-19021	62	36	,	,	PUNCT
bracis-19021	62	37	a.k.a	a.k.a	INTJ
bracis-19021	62	38	.	.	PUNCT
bracis-19021	63	1	the	the	DET
bracis-19021	63	2	dynamic	dynamic	ADJ
bracis-19021	63	3	and	and	CCONJ
bracis-19021	63	4	static	static	ADJ
bracis-19021	63	5	gains	gain	NOUN
bracis-19021	63	6	respectively	respectively	ADV
bracis-19021	63	7	in	in	ADP
bracis-19021	63	8	the	the	DET
bracis-19021	63	9	optimal	optimal	ADJ
bracis-19021	63	10	control	control	NOUN
bracis-19021	63	11	literature	literature	NOUN
bracis-19021	63	12	.	.	PUNCT
bracis-19021	64	1	here	here	ADV
bracis-19021	64	2	,	,	PUNCT
bracis-19021	64	3	\({\theta	\({\theta	PROPN
bracis-19021	64	4	}	}	PUNCT
bracis-19021	64	5	\	\	NOUN
bracis-19021	64	6	)	)	PUNCT
bracis-19021	64	7	is	be	AUX
bracis-19021	64	8	the	the	DET
bracis-19021	64	9	flattened	flatten	VERB
bracis-19021	64	10	parameter	parameter	NOUN
bracis-19021	64	11	vector	vector	NOUN
bracis-19021	64	12	corresponding	correspond	VERB
bracis-19021	64	13	to	to	ADP
bracis-19021	64	14	the	the	DET
bracis-19021	64	15	collection	collection	NOUN
bracis-19021	64	16	of	of	ADP
bracis-19021	64	17	function	function	NOUN
bracis-19021	64	18	coefficients	coefficient	NOUN
bracis-19021	64	19	.	.	PUNCT
bracis-19021	65	1	given	give	VERB
bracis-19021	65	2	any	any	DET
bracis-19021	65	3	linear	linear	ADJ
bracis-19021	65	4	policy	policy	NOUN
bracis-19021	65	5	\({\mu	\({\mu	ADP
bracis-19021	65	6	}	}	PUNCT
bracis-19021	65	7	_	_	PUNCT
bracis-19021	65	8	{	{	PUNCT
bracis-19021	65	9	{	{	PUNCT
bracis-19021	65	10	\theta	\theta	PROPN
bracis-19021	65	11	}	}	PUNCT
bracis-19021	65	12	}	}	PUNCT
bracis-19021	65	13	\	\	NOUN
bracis-19021	65	14	)	)	PUNCT
bracis-19021	65	15	,	,	PUNCT
bracis-19021	65	16	its	its	PRON
bracis-19021	65	17	state	state	NOUN
bracis-19021	65	18	-	-	PUNCT
bracis-19021	65	19	value	value	NOUN
bracis-19021	65	20	function	function	NOUN
bracis-19021	65	21	,	,	PUNCT
bracis-19021	65	22	the	the	DET
bracis-19021	65	23	expected	expect	VERB
bracis-19021	65	24	return	return	NOUN
bracis-19021	65	25	from	from	ADP
bracis-19021	65	26	each	each	DET
bracis-19021	65	27	state	state	NOUN
bracis-19021	65	28	,	,	PUNCT
bracis-19021	65	29	can	can	AUX
bracis-19021	65	30	be	be	AUX
bracis-19021	65	31	computed	compute	VERB
bracis-19021	65	32	recursively	recursively	ADV
bracis-19021	65	33	via	via	ADP
bracis-19021	65	34	the	the	DET
bracis-19021	65	35	bellman	bellman	NOUN
bracis-19021	65	36	equations	equation	NOUN
bracis-19021	66	1	[	[	X
bracis-19021	66	2	25	25	NUM
bracis-19021	66	3	]	]	X
bracis-19021	66	4	:	:	PUNCT
bracis-19021	66	5	(	(	PUNCT
bracis-19021	66	6	4	4	X
bracis-19021	66	7	)	)	PUNCT
bracis-19021	66	8	where	where	SCONJ
bracis-19021	66	9	\(v^{{\mu	\(v^{{\mu	NOUN
bracis-19021	66	10	}	}	PUNCT
bracis-19021	66	11	_	_	PUNCT
bracis-19021	66	12	{	{	PUNCT
bracis-19021	66	13	{	{	PUNCT
bracis-19021	66	14	\theta	\theta	PROPN
bracis-19021	66	15	}	}	PUNCT
bracis-19021	66	16	}	}	PUNCT
bracis-19021	66	17	}	}	PUNCT
bracis-19021	66	18	(	(	PUNCT
bracis-19021	66	19	{	{	PUNCT
bracis-19021	66	20	\mathbf	\mathbf	PROPN
bracis-19021	66	21	{	{	PUNCT
bracis-19021	66	22	s}}_{h	s}}_{h	NOUN
bracis-19021	66	23	}	}	PUNCT
bracis-19021	66	24	)	)	PUNCT
bracis-19021	66	25	=	=	SYM
bracis-19021	66	26	0\	0\	PROPN
bracis-19021	66	27	)	)	PUNCT
bracis-19021	66	28	and	and	CCONJ
bracis-19021	66	29	the	the	DET
bracis-19021	66	30	expectation	expectation	NOUN
bracis-19021	66	31	is	be	AUX
bracis-19021	66	32	w.r.t	w.r.t	ADJ
bracis-19021	66	33	.	.	PUNCT
bracis-19021	66	34	.	.	PUNCT
bracis-19021	67	1	since	since	SCONJ
bracis-19021	67	2	the	the	DET
bracis-19021	67	3	policy	policy	NOUN
bracis-19021	67	4	is	be	AUX
bracis-19021	67	5	linear	linear	ADJ
bracis-19021	67	6	,	,	PUNCT
bracis-19021	67	7	rewards	reward	NOUN
bracis-19021	67	8	are	be	AUX
bracis-19021	67	9	quadratic	quadratic	ADJ
bracis-19021	67	10	,	,	PUNCT
bracis-19021	67	11	and	and	CCONJ
bracis-19021	67	12	the	the	DET
bracis-19021	67	13	transitions	transition	NOUN
bracis-19021	67	14	,	,	PUNCT
bracis-19021	67	15	gaussian	gaussian	NOUN
bracis-19021	67	16	,	,	PUNCT
bracis-19021	67	17	the	the	DET
bracis-19021	67	18	expectations	expectation	NOUN
bracis-19021	67	19	in	in	ADP
bracis-19021	67	20	eq	eq	ADP
bracis-19021	67	21	.	.	PUNCT
bracis-19021	68	1	(	(	PUNCT
bracis-19021	68	2	4	4	X
bracis-19021	68	3	)	)	PUNCT
bracis-19021	68	4	can	can	AUX
bracis-19021	68	5	be	be	AUX
bracis-19021	68	6	computed	compute	VERB
bracis-19021	68	7	analytically	analytically	ADV
bracis-19021	68	8	by	by	ADP
bracis-19021	68	9	iteratively	iteratively	ADV
bracis-19021	68	10	solving	solve	VERB
bracis-19021	68	11	eq	eq	NOUN
bracis-19021	68	12	.	.	PUNCT
bracis-19021	69	1	(	(	PUNCT
bracis-19021	69	2	4	4	NUM
bracis-19021	69	3	)	)	PUNCT
bracis-19021	69	4	from	from	ADP
bracis-19021	69	5	timestep	timestep	NOUN
bracis-19021	69	6	\(h-1\	\(h-1\	PROPN
bracis-19021	69	7	)	)	PUNCT
bracis-19021	69	8	to	to	ADP
bracis-19021	69	9	0	0	NUM
bracis-19021	69	10	(	(	PUNCT
bracis-19021	69	11	a	a	DET
bracis-19021	69	12	dynamic	dynamic	ADJ
bracis-19021	69	13	programming	programming	NOUN
bracis-19021	69	14	method	method	NOUN
bracis-19021	69	15	)	)	PUNCT
bracis-19021	69	16	.	.	PUNCT
bracis-19021	70	1	the	the	DET
bracis-19021	70	2	solution	solution	NOUN
bracis-19021	70	3	\(v^{{\mu	\(v^{{\mu	NOUN
bracis-19021	70	4	}	}	PUNCT
bracis-19021	70	5	_	_	PUNCT
bracis-19021	70	6	{	{	PUNCT
bracis-19021	70	7	{	{	PUNCT
bracis-19021	70	8	\theta	\theta	PROPN
bracis-19021	70	9	}	}	PUNCT
bracis-19021	70	10	}	}	PUNCT
bracis-19021	70	11	}	}	PUNCT
bracis-19021	70	12	\	\	NOUN
bracis-19021	70	13	)	)	PUNCT
bracis-19021	70	14	is	be	AUX
bracis-19021	70	15	itself	itself	PRON
bracis-19021	70	16	quadratic	quadratic	ADJ
bracis-19021	70	17	and	and	CCONJ
bracis-19021	70	18	thus	thus	ADV
bracis-19021	70	19	the	the	DET
bracis-19021	70	20	policy	policy	NOUN
bracis-19021	70	21	return	return	NOUN
bracis-19021	70	22	can	can	AUX
bracis-19021	70	23	be	be	AUX
bracis-19021	70	24	computed	compute	VERB
bracis-19021	70	25	analytically	analytically	ADV
bracis-19021	70	26	:	:	PUNCT
bracis-19021	70	27	(	(	PUNCT
bracis-19021	70	28	5	5	NUM
bracis-19021	70	29	)	)	PUNCT
bracis-19021	70	30	moreover	moreover	ADV
bracis-19021	70	31	,	,	PUNCT
bracis-19021	70	32	lqgs	lqg	NOUN
bracis-19021	70	33	can	can	AUX
bracis-19021	70	34	be	be	AUX
bracis-19021	70	35	solved	solve	VERB
bracis-19021	70	36	for	for	ADP
bracis-19021	70	37	their	their	PRON
bracis-19021	70	38	optimal	optimal	ADJ
bracis-19021	70	39	policy	policy	NOUN
bracis-19021	70	40	\({\mu	\({\mu	ADP
bracis-19021	70	41	}	}	PUNCT
bracis-19021	70	42	_	_	PUNCT
bracis-19021	70	43	{	{	PUNCT
bracis-19021	70	44	{	{	PUNCT
bracis-19021	70	45	\theta	\theta	PROPN
bracis-19021	70	46	}	}	PUNCT
bracis-19021	70	47	}	}	PUNCT
bracis-19021	70	48	^\star	^\star	PUNCT
bracis-19021	70	49	\	\	NOUN
bracis-19021	70	50	)	)	PUNCT
bracis-19021	70	51	,	,	PUNCT
bracis-19021	70	52	which	which	PRON
bracis-19021	70	53	is	be	AUX
bracis-19021	70	54	time	time	NOUN
bracis-19021	70	55	-	-	PUNCT
bracis-19021	70	56	varying	vary	VERB
bracis-19021	70	57	linear	linear	NOUN
bracis-19021	70	58	,	,	PUNCT
bracis-19021	70	59	by	by	ADP
bracis-19021	70	60	modifying	modify	VERB
bracis-19021	70	61	the	the	DET
bracis-19021	70	62	dynamic	dynamic	ADJ
bracis-19021	70	63	programming	programming	NOUN
bracis-19021	70	64	method	method	NOUN
bracis-19021	70	65	above	above	ADV
bracis-19021	70	66	to	to	PART
bracis-19021	70	67	solve	solve	VERB
bracis-19021	70	68	for	for	ADP
bracis-19021	70	69	the	the	DET
bracis-19021	70	70	policy	policy	NOUN
bracis-19021	70	71	that	that	PRON
bracis-19021	70	72	maximizes	maximize	VERB
bracis-19021	70	73	the	the	DET
bracis-19021	70	74	expected	expect	VERB
bracis-19021	70	75	value	value	NOUN
bracis-19021	70	76	from	from	ADP
bracis-19021	70	77	each	each	DET
bracis-19021	70	78	timestep	timestep	NOUN
bracis-19021	70	79	based	base	VERB
bracis-19021	70	80	on	on	ADP
bracis-19021	70	81	eq	eq	ADP
bracis-19021	70	82	.	.	PUNCT
bracis-19021	71	1	(	(	PUNCT
bracis-19021	71	2	4	4	NUM
bracis-19021	71	3	)	)	PUNCT
bracis-19021	71	4	(	(	PUNCT
bracis-19021	71	5	also	also	ADV
bracis-19021	71	6	computable	computable	ADJ
bracis-19021	71	7	analytically	analytically	ADV
bracis-19021	71	8	)	)	PUNCT
bracis-19021	71	9	.	.	PUNCT
bracis-19021	72	1	algorithm	algorithm	PROPN
bracis-19021	72	2	 	 	SPACE
bracis-19021	72	3	1	1	NUM
bracis-19021	72	4	shows	show	VERB
bracis-19021	72	5	the	the	DET
bracis-19021	72	6	pseudocode	pseudocode	NOUN
bracis-19021	72	7	of	of	ADP
bracis-19021	72	8	the	the	DET
bracis-19021	72	9	procedure	procedure	NOUN
bracis-19021	72	10	we	we	PRON
bracis-19021	72	11	use	use	VERB
bracis-19021	72	12	to	to	PART
bracis-19021	72	13	derive	derive	VERB
bracis-19021	72	14	the	the	DET
bracis-19021	72	15	optimal	optimal	ADJ
bracis-19021	72	16	policy	policy	NOUN
bracis-19021	72	17	,	,	PUNCT
bracis-19021	72	18	which	which	PRON
bracis-19021	72	19	can	can	AUX
bracis-19021	72	20	be	be	AUX
bracis-19021	72	21	slightly	slightly	ADV
bracis-19021	72	22	modified	modify	VERB
bracis-19021	72	23	to	to	PART
bracis-19021	72	24	return	return	VERB
bracis-19021	72	25	the	the	DET
bracis-19021	72	26	value	value	NOUN
bracis-19021	72	27	function	function	NOUN
bracis-19021	72	28	for	for	ADP
bracis-19021	72	29	a	a	DET
bracis-19021	72	30	given	give	VERB
bracis-19021	72	31	policy	policy	NOUN
bracis-19021	72	32	.	.	PUNCT
bracis-19021	73	1	lqg	lqg	NOUN
bracis-19021	73	2	is	be	AUX
bracis-19021	73	3	thus	thus	ADV
bracis-19021	73	4	one	one	NUM
bracis-19021	73	5	of	of	ADP
bracis-19021	73	6	the	the	DET
bracis-19021	73	7	few	few	ADJ
bracis-19021	73	8	classes	class	NOUN
bracis-19021	73	9	of	of	ADP
bracis-19021	73	10	nontrivial	nontrivial	ADJ
bracis-19021	73	11	continuous	continuous	ADJ
bracis-19021	73	12	control	control	NOUN
bracis-19021	73	13	problems	problem	NOUN
bracis-19021	73	14	which	which	PRON
bracis-19021	73	15	allows	allow	VERB
bracis-19021	73	16	us	we	PRON
bracis-19021	73	17	to	to	PART
bracis-19021	73	18	evaluate	evaluate	VERB
bracis-19021	73	19	rl	rl	ADP
bracis-19021	73	20	methods	method	NOUN
bracis-19021	73	21	against	against	ADP
bracis-19021	73	22	the	the	DET
bracis-19021	73	23	theoretical	theoretical	ADJ
bracis-19021	73	24	best	good	ADJ
bracis-19021	73	25	solutions	solution	NOUN
bracis-19021	73	26	.	.	PUNCT
bracis-19021	74	1	3.3	3.3	NUM
bracis-19021	74	2	stochastic	stochastic	ADJ
bracis-19021	74	3	value	value	NOUN
bracis-19021	74	4	gradient	gradient	NOUN
bracis-19021	74	5	methods	method	NOUN
bracis-19021	74	6	in	in	ADP
bracis-19021	74	7	the	the	DET
bracis-19021	74	8	broader	broad	ADJ
bracis-19021	74	9	rl	rl	AUX
bracis-19021	74	10	context	context	NOUN
bracis-19021	74	11	,	,	PUNCT
bracis-19021	74	12	methods	method	NOUN
bracis-19021	74	13	that	that	PRON
bracis-19021	74	14	learn	learn	VERB
bracis-19021	74	15	parameterized	parameterized	ADJ
bracis-19021	74	16	policies	policy	NOUN
bracis-19021	74	17	,	,	PUNCT
bracis-19021	74	18	often	often	ADV
bracis-19021	74	19	called	call	VERB
bracis-19021	74	20	policy	policy	NOUN
bracis-19021	74	21	optimization	optimization	NOUN
bracis-19021	74	22	methods	method	NOUN
bracis-19021	74	23	,	,	PUNCT
bracis-19021	74	24	have	have	AUX
bracis-19021	74	25	gained	gain	VERB
bracis-19021	74	26	traction	traction	NOUN
bracis-19021	74	27	in	in	ADP
bracis-19021	74	28	the	the	DET
bracis-19021	74	29	recent	recent	ADJ
bracis-19021	74	30	decade	decade	NOUN
bracis-19021	74	31	.	.	PUNCT
bracis-19021	75	1	as	as	ADP
bracis-19021	75	2	function	function	NOUN
bracis-19021	75	3	approximation	approximation	NOUN
bracis-19021	75	4	research	research	NOUN
bracis-19021	75	5	,	,	PUNCT
bracis-19021	75	6	specially	specially	ADV
bracis-19021	75	7	on	on	ADP
bracis-19021	75	8	deep	deep	ADJ
bracis-19021	75	9	learning	learning	NOUN
bracis-19021	75	10	,	,	PUNCT
bracis-19021	75	11	has	have	AUX
bracis-19021	75	12	advanced	advance	VERB
bracis-19021	75	13	,	,	PUNCT
bracis-19021	75	14	parameterized	parameterized	ADJ
bracis-19021	75	15	policies	policy	NOUN
bracis-19021	75	16	were	be	AUX
bracis-19021	75	17	able	able	ADJ
bracis-19021	75	18	to	to	PART
bracis-19021	75	19	unify	unify	VERB
bracis-19021	75	20	perception	perception	NOUN
bracis-19021	75	21	(	(	PUNCT
bracis-19021	75	22	processing	process	VERB
bracis-19021	75	23	sensorial	sensorial	ADJ
bracis-19021	75	24	input	input	NOUN
bracis-19021	75	25	from	from	ADP
bracis-19021	75	26	the	the	DET
bracis-19021	75	27	environment	environment	NOUN
bracis-19021	75	28	)	)	PUNCT
bracis-19021	75	29	and	and	CCONJ
bracis-19021	75	30	decision	decision	NOUN
bracis-19021	75	31	-	-	PUNCT
bracis-19021	75	32	making	making	NOUN
bracis-19021	75	33	(	(	PUNCT
bracis-19021	75	34	choosing	choose	VERB
bracis-19021	75	35	actions	action	NOUN
bracis-19021	75	36	to	to	PART
bracis-19021	75	37	maximize	maximize	VERB
bracis-19021	75	38	return	return	NOUN
bracis-19021	75	39	)	)	PUNCT
bracis-19021	75	40	tasks	task	NOUN
bracis-19021	75	41	[	[	X
bracis-19021	75	42	15	15	NUM
bracis-19021	75	43	]	]	PUNCT
bracis-19021	75	44	.	.	PUNCT
bracis-19021	76	1	to	to	PART
bracis-19021	76	2	improve	improve	VERB
bracis-19021	76	3	such	such	ADJ
bracis-19021	76	4	parameterized	parameterized	ADJ
bracis-19021	76	5	approximators	approximator	NOUN
bracis-19021	76	6	from	from	ADP
bracis-19021	76	7	data	datum	NOUN
bracis-19021	76	8	,	,	PUNCT
bracis-19021	76	9	the	the	DET
bracis-19021	76	10	workhorse	workhorse	NOUN
bracis-19021	76	11	behind	behind	ADP
bracis-19021	76	12	many	many	ADJ
bracis-19021	76	13	policy	policy	NOUN
bracis-19021	76	14	optimization	optimization	NOUN
bracis-19021	76	15	methods	method	NOUN
bracis-19021	76	16	is	be	AUX
bracis-19021	76	17	stochastic	stochastic	ADJ
bracis-19021	76	18	gradient	gradient	ADJ
bracis-19021	76	19	descent	descent	NOUN
bracis-19021	76	20	(	(	PUNCT
bracis-19021	76	21	sgd	sgd	PROPN
bracis-19021	76	22	)	)	PUNCT
bracis-19021	77	1	[	[	X
bracis-19021	77	2	22	22	NUM
bracis-19021	77	3	]	]	PUNCT
bracis-19021	77	4	.	.	PUNCT
bracis-19021	78	1	thus	thus	ADV
bracis-19021	78	2	,	,	PUNCT
bracis-19021	78	3	it	it	PRON
bracis-19021	78	4	is	be	AUX
bracis-19021	78	5	imperative	imperative	ADJ
bracis-19021	78	6	to	to	PART
bracis-19021	78	7	estimate	estimate	VERB
bracis-19021	78	8	the	the	DET
bracis-19021	78	9	gradient	gradient	NOUN
bracis-19021	78	10	of	of	ADP
bracis-19021	78	11	the	the	DET
bracis-19021	78	12	expected	expect	VERB
bracis-19021	78	13	return	return	NOUN
bracis-19021	78	14	w.r.t	w.r.t	VERB
bracis-19021	78	15	.	.	PUNCT
bracis-19021	79	1	policy	policy	NOUN
bracis-19021	79	2	parameters	parameter	NOUN
bracis-19021	79	3	,	,	PUNCT
bracis-19021	79	4	a.k.a	a.k.a	INTJ
bracis-19021	79	5	.	.	PUNCT
bracis-19021	80	1	the	the	DET
bracis-19021	80	2	value	value	NOUN
bracis-19021	80	3	gradient	gradient	NOUN
bracis-19021	80	4	,	,	PUNCT
bracis-19021	80	5	from	from	ADP
bracis-19021	80	6	data	datum	NOUN
bracis-19021	80	7	(	(	PUNCT
bracis-19021	80	8	states	state	NOUN
bracis-19021	80	9	,	,	PUNCT
bracis-19021	80	10	actions	action	NOUN
bracis-19021	80	11	and	and	CCONJ
bracis-19021	80	12	rewards	reward	NOUN
bracis-19021	80	13	)	)	PUNCT
bracis-19021	80	14	collected	collect	VERB
bracis-19021	80	15	via	via	ADP
bracis-19021	80	16	interaction	interaction	NOUN
bracis-19021	80	17	with	with	ADP
bracis-19021	80	18	the	the	DET
bracis-19021	80	19	environment	environment	NOUN
bracis-19021	80	20	.	.	PUNCT
bracis-19021	81	1	svg	svg	NOUN
bracis-19021	81	2	methods	method	NOUN
bracis-19021	81	3	build	build	VERB
bracis-19021	81	4	gradient	gradient	ADJ
bracis-19021	81	5	estimates	estimate	NOUN
bracis-19021	81	6	by	by	ADP
bracis-19021	81	7	first	first	ADV
bracis-19021	81	8	using	use	VERB
bracis-19021	81	9	the	the	DET
bracis-19021	81	10	available	available	ADJ
bracis-19021	81	11	data	datum	NOUN
bracis-19021	81	12	to	to	PART
bracis-19021	81	13	learn	learn	VERB
bracis-19021	81	14	a	a	DET
bracis-19021	81	15	model	model	NOUN
bracis-19021	81	16	of	of	ADP
bracis-19021	81	17	the	the	DET
bracis-19021	81	18	environment	environment	NOUN
bracis-19021	81	19	,	,	PUNCT
bracis-19021	81	20	i.e.	i.e.	X
bracis-19021	81	21	,	,	PUNCT
bracis-19021	81	22	a	a	DET
bracis-19021	81	23	function	function	NOUN
bracis-19021	81	24	approximator	approximator	NOUN
bracis-19021	81	25	.	.	PUNCT
bracis-19021	82	1	a	a	DET
bracis-19021	82	2	common	common	ADJ
bracis-19021	82	3	approach	approach	NOUN
bracis-19021	82	4	to	to	ADP
bracis-19021	82	5	leveraging	leverage	VERB
bracis-19021	82	6	the	the	DET
bracis-19021	82	7	learned	learn	VERB
bracis-19021	82	8	model	model	NOUN
bracis-19021	82	9	is	be	AUX
bracis-19021	82	10	as	as	SCONJ
bracis-19021	82	11	follows	follow	VERB
bracis-19021	82	12	.	.	PUNCT
bracis-19021	83	1	first	first	ADV
bracis-19021	83	2	,	,	PUNCT
bracis-19021	83	3	the	the	DET
bracis-19021	83	4	agent	agent	NOUN
bracis-19021	83	5	collects	collect	VERB
bracis-19021	83	6	b	b	PROPN
bracis-19021	83	7	states	state	NOUN
bracis-19021	83	8	via	via	ADP
bracis-19021	83	9	interaction	interaction	NOUN
bracis-19021	83	10	with	with	ADP
bracis-19021	83	11	the	the	DET
bracis-19021	83	12	environment	environment	NOUN
bracis-19021	83	13	,	,	PUNCT
bracis-19021	83	14	potentially	potentially	ADV
bracis-19021	83	15	with	with	ADP
bracis-19021	83	16	an	an	DET
bracis-19021	83	17	exploratory	exploratory	ADJ
bracis-19021	83	18	policy	policy	NOUN
bracis-19021	83	19	\(\beta	\(\beta	NOUN
bracis-19021	83	20	\	\	PROPN
bracis-19021	83	21	)	)	PUNCT
bracis-19021	83	22	(	(	PUNCT
bracis-19021	83	23	we	we	PRON
bracis-19021	83	24	use	use	VERB
bracis-19021	83	25	\({\mathbf	\({\mathbf	NOUN
bracis-19021	83	26	{	{	PUNCT
bracis-19021	83	27	s}}_t	s}}_t	NOUN
bracis-19021	83	28	\sim	\sim	NOUN
bracis-19021	83	29	d^{\beta	d^{\beta	PROPN
bracis-19021	83	30	}	}	PUNCT
bracis-19021	83	31	\	\	NOUN
bracis-19021	83	32	)	)	PUNCT
bracis-19021	83	33	to	to	PART
bracis-19021	83	34	denote	denote	VERB
bracis-19021	83	35	sampling	sample	VERB
bracis-19021	83	36	from	from	ADP
bracis-19021	83	37	its	its	PRON
bracis-19021	83	38	induced	induced	ADJ
bracis-19021	83	39	state	state	NOUN
bracis-19021	83	40	distribution	distribution	NOUN
bracis-19021	83	41	)	)	PUNCT
bracis-19021	83	42	.	.	PUNCT
bracis-19021	84	1	then	then	ADV
bracis-19021	84	2	,	,	PUNCT
bracis-19021	84	3	it	it	PRON
bracis-19021	84	4	generates	generate	VERB
bracis-19021	84	5	short	short	ADJ
bracis-19021	84	6	model	model	NOUN
bracis-19021	84	7	-	-	PUNCT
bracis-19021	84	8	based	base	VERB
bracis-19021	84	9	trajectories	trajectory	NOUN
bracis-19021	84	10	with	with	ADP
bracis-19021	84	11	the	the	DET
bracis-19021	84	12	current	current	ADJ
bracis-19021	84	13	policy	policy	NOUN
bracis-19021	84	14	\({\mu	\({\mu	ADP
bracis-19021	84	15	}	}	PUNCT
bracis-19021	84	16	_	_	PUNCT
bracis-19021	84	17	{	{	PUNCT
bracis-19021	84	18	{	{	PUNCT
bracis-19021	84	19	\theta	\theta	PROPN
bracis-19021	84	20	}	}	PUNCT
bracis-19021	84	21	}	}	PUNCT
bracis-19021	84	22	\	\	NOUN
bracis-19021	84	23	)	)	PUNCT
bracis-19021	84	24	,	,	PUNCT
bracis-19021	84	25	branching	branch	VERB
bracis-19021	84	26	off	off	ADP
bracis-19021	84	27	the	the	DET
bracis-19021	84	28	states	state	NOUN
bracis-19021	84	29	previous	previous	ADJ
bracis-19021	84	30	collected	collect	VERB
bracis-19021	84	31	.	.	PUNCT
bracis-19021	85	1	finally	finally	ADV
bracis-19021	85	2	,	,	PUNCT
bracis-19021	85	3	it	it	PRON
bracis-19021	85	4	computes	compute	VERB
bracis-19021	85	5	the	the	DET
bracis-19021	85	6	average	average	ADJ
bracis-19021	85	7	model	model	NOUN
bracis-19021	85	8	-	-	PUNCT
bracis-19021	85	9	based	base	VERB
bracis-19021	85	10	returns	return	NOUN
bracis-19021	85	11	and	and	CCONJ
bracis-19021	85	12	forms	form	NOUN
bracis-19021	85	13	an	an	DET
bracis-19021	85	14	estimate	estimate	NOUN
bracis-19021	85	15	of	of	ADP
bracis-19021	85	16	the	the	DET
bracis-19021	85	17	value	value	NOUN
bracis-19021	85	18	gradient	gradient	NOUN
bracis-19021	85	19	using	use	VERB
bracis-19021	85	20	backpropagation	backpropagation	NOUN
bracis-19021	85	21	[	[	X
bracis-19021	85	22	7	7	NUM
bracis-19021	85	23	]	]	SYM
bracis-19021	85	24	:	:	PUNCT
bracis-19021	85	25	(	(	PUNCT
bracis-19021	85	26	6	6	NUM
bracis-19021	85	27	)	)	PUNCT
bracis-19021	85	28	here	here	ADV
bracis-19021	85	29	,	,	PUNCT
bracis-19021	85	30	\(\hat{q}^{{\mu	\(\hat{q}^{{\mu	ADV
bracis-19021	85	31	}	}	PUNCT
bracis-19021	85	32	_	_	PUNCT
bracis-19021	85	33	{	{	PUNCT
bracis-19021	85	34	{	{	PUNCT
bracis-19021	85	35	\theta	\theta	PROPN
bracis-19021	85	36	}	}	PUNCT
bracis-19021	85	37	}	}	PUNCT
bracis-19021	85	38	}	}	PUNCT
bracis-19021	85	39	\	\	NOUN
bracis-19021	85	40	)	)	PUNCT
bracis-19021	85	41	is	be	AUX
bracis-19021	85	42	an	an	DET
bracis-19021	85	43	approximation	approximation	NOUN
bracis-19021	85	44	(	(	PUNCT
bracis-19021	85	45	e.g.	e.g.	ADV
bracis-19021	85	46	,	,	PUNCT
bracis-19021	85	47	a	a	DET
bracis-19021	85	48	learned	learn	VERB
bracis-19021	85	49	neural	neural	ADJ
bracis-19021	85	50	network	network	NOUN
bracis-19021	85	51	)	)	PUNCT
bracis-19021	85	52	of	of	ADP
bracis-19021	85	53	the	the	DET
bracis-19021	85	54	policy	policy	NOUN
bracis-19021	85	55	’s	’s	PART
bracis-19021	85	56	action	action	NOUN
bracis-19021	85	57	-	-	PUNCT
bracis-19021	85	58	value	value	NOUN
bracis-19021	85	59	function	function	NOUN
bracis-19021	85	60	.	.	PUNCT
bracis-19021	86	1	we	we	PRON
bracis-19021	86	2	refer	refer	VERB
bracis-19021	86	3	to	to	ADP
bracis-19021	86	4	eq	eq	NOUN
bracis-19021	86	5	.	.	PUNCT
bracis-19021	87	1	(	(	PUNCT
bracis-19021	87	2	6	6	NUM
bracis-19021	87	3	)	)	PUNCT
bracis-19021	87	4	as	as	ADP
bracis-19021	87	5	the	the	DET
bracis-19021	87	6	maac(k	maac(k	NOUN
bracis-19021	87	7	)	)	PUNCT
bracis-19021	87	8	estimator	estimator	NOUN
bracis-19021	87	9	,	,	PUNCT
bracis-19021	87	10	as	as	SCONJ
bracis-19021	87	11	it	it	PRON
bracis-19021	87	12	uses	use	VERB
bracis-19021	87	13	k	k	PROPN
bracis-19021	87	14	steps	step	NOUN
bracis-19021	87	15	of	of	ADP
bracis-19021	87	16	simulated	simulated	ADJ
bracis-19021	87	17	interaction	interaction	NOUN
bracis-19021	87	18	and	and	CCONJ
bracis-19021	87	19	was	be	AUX
bracis-19021	87	20	featured	feature	VERB
bracis-19021	87	21	prominently	prominently	ADV
bracis-19021	87	22	in	in	ADP
bracis-19021	87	23	the	the	DET
bracis-19021	87	24	maac	maac	PROPN
bracis-19021	87	25	paper	paper	NOUN
bracis-19021	88	1	[	[	X
bracis-19021	88	2	4].footnote	4].footnote	NUM
bracis-19021	88	3	1	1	NUM
bracis-19021	88	4	we	we	PRON
bracis-19021	88	5	question	question	VERB
bracis-19021	88	6	,	,	PUNCT
bracis-19021	88	7	however	however	ADV
bracis-19021	88	8	,	,	PUNCT
bracis-19021	88	9	if	if	SCONJ
bracis-19021	88	10	eq	eq	ADJ
bracis-19021	88	11	.	.	PUNCT
bracis-19021	88	12	(	(	PUNCT
bracis-19021	88	13	6	6	NUM
bracis-19021	88	14	)	)	PUNCT
bracis-19021	88	15	actually	actually	ADV
bracis-19021	88	16	provides	provide	VERB
bracis-19021	88	17	good	good	ADJ
bracis-19021	88	18	empirical	empirical	ADJ
bracis-19021	88	19	estimates	estimate	NOUN
bracis-19021	88	20	of	of	ADP
bracis-19021	88	21	the	the	DET
bracis-19021	88	22	true	true	ADJ
bracis-19021	88	23	value	value	NOUN
bracis-19021	88	24	gradient	gradient	NOUN
bracis-19021	88	25	.	.	PUNCT
bracis-19021	89	1	to	to	PART
bracis-19021	89	2	elucidate	elucidate	VERB
bracis-19021	89	3	this	this	DET
bracis-19021	89	4	matter	matter	NOUN
bracis-19021	89	5	,	,	PUNCT
bracis-19021	89	6	we	we	PRON
bracis-19021	89	7	compare	compare	VERB
bracis-19021	89	8	maac(k	maac(k	NOUN
bracis-19021	89	9	)	)	PUNCT
bracis-19021	89	10	to	to	ADP
bracis-19021	89	11	the	the	DET
bracis-19021	89	12	value	value	NOUN
bracis-19021	89	13	gradient	gradient	NOUN
bracis-19021	89	14	estimator	estimator	NOUN
bracis-19021	89	15	provided	provide	VERB
bracis-19021	89	16	by	by	ADP
bracis-19021	89	17	the	the	DET
bracis-19021	89	18	dpg	dpg	PROPN
bracis-19021	89	19	theorem	theorem	NOUN
bracis-19021	89	20	[	[	X
bracis-19021	89	21	24	24	NUM
bracis-19021	89	22	]	]	NUM
bracis-19021	89	23	:	:	PUNCT
bracis-19021	89	24	(	(	PUNCT
bracis-19021	89	25	7	7	X
bracis-19021	89	26	)	)	PUNCT
bracis-19021	89	27	besides	besides	SCONJ
bracis-19021	89	28	the	the	DET
bracis-19021	89	29	fact	fact	NOUN
bracis-19021	89	30	that	that	SCONJ
bracis-19021	89	31	eq	eq	X
bracis-19021	89	32	.	.	PUNCT
bracis-19021	90	1	(	(	PUNCT
bracis-19021	90	2	7	7	X
bracis-19021	90	3	)	)	PUNCT
bracis-19021	90	4	requires	require	VERB
bracis-19021	90	5	us	we	PRON
bracis-19021	90	6	to	to	PART
bracis-19021	90	7	use	use	VERB
bracis-19021	90	8	the	the	DET
bracis-19021	90	9	on	on	ADP
bracis-19021	90	10	-	-	PUNCT
bracis-19021	90	11	policy	policy	NOUN
bracis-19021	90	12	distribution	distribution	NOUN
bracis-19021	90	13	of	of	ADP
bracis-19021	90	14	states	state	NOUN
bracis-19021	90	15	\(d^{{\mu	\(d^{{\mu	VERB
bracis-19021	90	16	}	}	PUNCT
bracis-19021	90	17	_	_	PUNCT
bracis-19021	90	18	{	{	PUNCT
bracis-19021	90	19	{	{	PUNCT
bracis-19021	90	20	\theta	\theta	PROPN
bracis-19021	90	21	}	}	PUNCT
bracis-19021	90	22	}	}	PUNCT
bracis-19021	90	23	}	}	PUNCT
bracis-19021	90	24	\	\	NOUN
bracis-19021	90	25	)	)	PUNCT
bracis-19021	90	26	,	,	PUNCT
bracis-19021	90	27	more	more	ADV
bracis-19021	90	28	subtle	subtle	ADJ
bracis-19021	90	29	differences	difference	NOUN
bracis-19021	90	30	with	with	ADP
bracis-19021	90	31	eq	eq	NOUN
bracis-19021	90	32	.	.	PUNCT
bracis-19021	91	1	(	(	PUNCT
bracis-19021	91	2	6	6	NUM
bracis-19021	91	3	)	)	PUNCT
bracis-19021	91	4	can	can	AUX
bracis-19021	91	5	be	be	AUX
bracis-19021	91	6	seen	see	VERB
bracis-19021	91	7	by	by	ADP
bracis-19021	91	8	expanding	expand	VERB
bracis-19021	91	9	the	the	DET
bracis-19021	91	10	definition	definition	NOUN
bracis-19021	91	11	of	of	ADP
bracis-19021	91	12	the	the	DET
bracis-19021	91	13	action	action	NOUN
bracis-19021	91	14	-	-	PUNCT
bracis-19021	91	15	value	value	NOUN
bracis-19021	91	16	function	function	NOUN
bracis-19021	91	17	to	to	PART
bracis-19021	91	18	form	form	VERB
bracis-19021	91	19	a	a	DET
bracis-19021	91	20	k	k	ADJ
bracis-19021	91	21	-	-	PUNCT
bracis-19021	91	22	step	step	NOUN
bracis-19021	91	23	version	version	NOUN
bracis-19021	91	24	of	of	ADP
bracis-19021	91	25	eq	eq	PROPN
bracis-19021	91	26	.	.	PUNCT
bracis-19021	92	1	(	(	PUNCT
bracis-19021	92	2	7	7	NUM
bracis-19021	92	3	)	)	PUNCT
bracis-19021	92	4	,	,	PUNCT
bracis-19021	92	5	which	which	PRON
bracis-19021	92	6	we	we	PRON
bracis-19021	92	7	call	call	VERB
bracis-19021	92	8	the	the	DET
bracis-19021	92	9	dpg(k	dpg(k	NOUN
bracis-19021	92	10	)	)	PUNCT
bracis-19021	92	11	estimator	estimator	NOUN
bracis-19021	92	12	.	.	PUNCT
bracis-19021	92	13	note	note	VERB
bracis-19021	92	14	that	that	SCONJ
bracis-19021	92	15	dpg(0	dpg(0	NOUN
bracis-19021	92	16	)	)	PUNCT
bracis-19021	92	17	is	be	AUX
bracis-19021	92	18	equivalent	equivalent	ADJ
bracis-19021	92	19	to	to	ADP
bracis-19021	92	20	eq	eq	PROPN
bracis-19021	92	21	.	.	PUNCT
bracis-19021	93	1	(	(	PUNCT
bracis-19021	93	2	7	7	NUM
bracis-19021	93	3	)	)	PUNCT
bracis-19021	93	4	.	.	PUNCT
bracis-19021	94	1	figure	figure	NOUN
bracis-19021	94	2	 	 	SPACE
bracis-19021	94	3	1	1	NUM
bracis-19021	94	4	shows	show	VERB
bracis-19021	94	5	the	the	DET
bracis-19021	94	6	stochastic	stochastic	ADJ
bracis-19021	94	7	computation	computation	NOUN
bracis-19021	94	8	graphs	graph	NOUN
bracis-19021	94	9	(	(	PUNCT
bracis-19021	94	10	scgs	scgs	PROPN
bracis-19021	94	11	)	)	PUNCT
bracis-19021	94	12	of	of	ADP
bracis-19021	94	13	the	the	DET
bracis-19021	94	14	maac(k	maac(k	NOUN
bracis-19021	94	15	)	)	PUNCT
bracis-19021	94	16	and	and	CCONJ
bracis-19021	94	17	dpg(k	dpg(k	PROPN
bracis-19021	94	18	)	)	PUNCT
bracis-19021	94	19	estimators	estimator	NOUN
bracis-19021	94	20	[	[	X
bracis-19021	94	21	23	23	NUM
bracis-19021	94	22	]	]	PUNCT
bracis-19021	94	23	.	.	PUNCT
bracis-19021	95	1	here	here	ADV
bracis-19021	95	2	,	,	PUNCT
bracis-19021	95	3	borderless	borderless	NOUN
bracis-19021	95	4	nodes	node	NOUN
bracis-19021	95	5	denote	denote	VERB
bracis-19021	95	6	input	input	NOUN
bracis-19021	95	7	variables	variable	NOUN
bracis-19021	95	8	;	;	PUNCT
bracis-19021	95	9	circles	circle	NOUN
bracis-19021	95	10	denote	denote	VERB
bracis-19021	95	11	stochastic	stochastic	ADJ
bracis-19021	95	12	nodes	node	NOUN
bracis-19021	95	13	,	,	PUNCT
bracis-19021	95	14	distributed	distribute	VERB
bracis-19021	95	15	conditionally	conditionally	ADV
bracis-19021	95	16	on	on	ADP
bracis-19021	95	17	their	their	PRON
bracis-19021	95	18	parents	parent	NOUN
bracis-19021	95	19	(	(	PUNCT
bracis-19021	95	20	if	if	SCONJ
bracis-19021	95	21	any	any	PRON
bracis-19021	95	22	)	)	PUNCT
bracis-19021	95	23	;	;	PUNCT
bracis-19021	95	24	and	and	CCONJ
bracis-19021	95	25	squares	square	NOUN
bracis-19021	95	26	denote	denote	VERB
bracis-19021	95	27	deterministic	deterministic	ADJ
bracis-19021	95	28	nodes	node	NOUN
bracis-19021	95	29	,	,	PUNCT
bracis-19021	95	30	which	which	PRON
bracis-19021	95	31	are	be	AUX
bracis-19021	95	32	functions	function	NOUN
bracis-19021	95	33	of	of	ADP
bracis-19021	95	34	their	their	PRON
bracis-19021	95	35	parents	parent	NOUN
bracis-19021	95	36	.	.	PUNCT
bracis-19021	96	1	because	because	SCONJ
bracis-19021	96	2	of	of	ADP
bracis-19021	96	3	the	the	DET
bracis-19021	96	4	term	term	NOUN
bracis-19021	96	5	in	in	ADP
bracis-19021	96	6	eq	eq	ADP
bracis-19021	96	7	.	.	PUNCT
bracis-19021	96	8	(	(	PUNCT
bracis-19021	96	9	7	7	NUM
bracis-19021	96	10	)	)	PUNCT
bracis-19021	96	11	,	,	PUNCT
bracis-19021	96	12	we	we	PRON
bracis-19021	96	13	’re	’re	AUX
bracis-19021	96	14	not	not	PART
bracis-19021	96	15	allowed	allow	VERB
bracis-19021	96	16	to	to	PART
bracis-19021	96	17	compute	compute	VERB
bracis-19021	96	18	the	the	DET
bracis-19021	96	19	gradients	gradient	NOUN
bracis-19021	96	20	of	of	ADP
bracis-19021	96	21	future	future	ADJ
bracis-19021	96	22	actions	action	NOUN
bracis-19021	96	23	w.r.t	w.r.t	VERB
bracis-19021	96	24	.	.	PUNCT
bracis-19021	97	1	policy	policy	NOUN
bracis-19021	97	2	parameters	parameter	NOUN
bracis-19021	97	3	in	in	ADP
bracis-19021	97	4	dpg(k	dpg(k	NOUN
bracis-19021	97	5	)	)	PUNCT
bracis-19021	97	6	,	,	PUNCT
bracis-19021	97	7	hence	hence	ADV
bracis-19021	97	8	why	why	SCONJ
bracis-19021	97	9	only	only	ADV
bracis-19021	97	10	the	the	DET
bracis-19021	97	11	first	first	ADJ
bracis-19021	97	12	action	action	NOUN
bracis-19021	97	13	has	have	VERB
bracis-19021	97	14	a	a	DET
bracis-19021	97	15	link	link	NOUN
bracis-19021	97	16	with	with	ADP
bracis-19021	97	17	\({\theta	\({\theta	PROPN
bracis-19021	97	18	}	}	PUNCT
bracis-19021	97	19	\	\	PROPN
bracis-19021	97	20	)	)	PUNCT
bracis-19021	97	21	.	.	PUNCT
bracis-19021	98	1	on	on	ADP
bracis-19021	98	2	the	the	DET
bracis-19021	98	3	other	other	ADJ
bracis-19021	98	4	hand	hand	NOUN
bracis-19021	98	5	,	,	PUNCT
bracis-19021	98	6	maac(k	maac(k	PROPN
bracis-19021	98	7	)	)	PUNCT
bracis-19021	98	8	backpropagates	backpropagate	VERB
bracis-19021	98	9	the	the	DET
bracis-19021	98	10	gradients	gradient	NOUN
bracis-19021	98	11	of	of	ADP
bracis-19021	98	12	the	the	DET
bracis-19021	98	13	rewards	reward	NOUN
bracis-19021	98	14	and	and	CCONJ
bracis-19021	98	15	value	value	NOUN
bracis-19021	98	16	-	-	PUNCT
bracis-19021	98	17	function	function	NOUN
bracis-19021	98	18	through	through	ADP
bracis-19021	98	19	all	all	DET
bracis-19021	98	20	intermediate	intermediate	ADJ
bracis-19021	98	21	actions	action	NOUN
bracis-19021	98	22	.	.	PUNCT
bracis-19021	99	1	our	our	PRON
bracis-19021	99	2	work	work	NOUN
bracis-19021	99	3	aims	aim	VERB
bracis-19021	99	4	at	at	ADP
bracis-19021	99	5	identifying	identify	VERB
bracis-19021	99	6	the	the	DET
bracis-19021	99	7	practical	practical	ADJ
bracis-19021	99	8	implications	implication	NOUN
bracis-19021	99	9	of	of	ADP
bracis-19021	99	10	these	these	DET
bracis-19021	99	11	differences	difference	NOUN
bracis-19021	99	12	and	and	CCONJ
bracis-19021	99	13	perhaps	perhaps	ADV
bracis-19021	99	14	help	help	VERB
bracis-19021	99	15	explain	explain	VERB
bracis-19021	99	16	why	why	SCONJ
bracis-19021	99	17	maac(k	maac(k	NOUN
bracis-19021	99	18	)	)	PUNCT
bracis-19021	99	19	has	have	AUX
bracis-19021	99	20	been	be	AUX
bracis-19021	99	21	used	use	VERB
bracis-19021	99	22	in	in	ADP
bracis-19021	99	23	svg	svg	NOUN
bracis-19021	99	24	methods	method	NOUN
bracis-19021	99	25	and	and	CCONJ
bracis-19021	99	26	not	not	PART
bracis-19021	99	27	dpg(k	dpg(k	NOUN
bracis-19021	99	28	)	)	PUNCT
bracis-19021	99	29	.	.	PUNCT
bracis-19021	100	1	fig	fig	NOUN
bracis-19021	100	2	.	.	PUNCT
bracis-19021	101	1	1	1	X
bracis-19021	101	2	.	.	X
bracis-19021	101	3	stochastic	stochastic	ADJ
bracis-19021	101	4	computation	computation	NOUN
bracis-19021	101	5	graphs	graph	NOUN
bracis-19021	101	6	for	for	ADP
bracis-19021	101	7	policy	policy	NOUN
bracis-19021	101	8	gradients	gradient	NOUN
bracis-19021	101	9	.	.	PUNCT
bracis-19021	102	1	(	(	PUNCT
bracis-19021	102	2	left	leave	VERB
bracis-19021	102	3	)	)	PUNCT
bracis-19021	102	4	model	model	NOUN
bracis-19021	102	5	-	-	PUNCT
bracis-19021	102	6	free	free	ADJ
bracis-19021	102	7	dpg	dpg	PROPN
bracis-19021	102	8	.	.	PUNCT
bracis-19021	103	1	(	(	PUNCT
bracis-19021	103	2	right	right	ADJ
bracis-19021	103	3	)	)	PUNCT
bracis-19021	103	4	model	model	NOUN
bracis-19021	103	5	-	-	PUNCT
bracis-19021	103	6	based	base	VERB
bracis-19021	103	7	dpg	dpg	PROPN
bracis-19021	103	8	:	:	PUNCT
bracis-19021	103	9	the	the	DET
bracis-19021	103	10	dashed	dash	VERB
bracis-19021	103	11	edges	edge	NOUN
bracis-19021	103	12	represent	represent	VERB
bracis-19021	103	13	the	the	DET
bracis-19021	103	14	k	k	ADJ
bracis-19021	103	15	-	-	PUNCT
bracis-19021	103	16	step	step	NOUN
bracis-19021	103	17	deterministic	deterministic	ADJ
bracis-19021	103	18	dependencies	dependency	NOUN
bracis-19021	103	19	of	of	ADP
bracis-19021	103	20	the	the	DET
bracis-19021	103	21	policy	policy	NOUN
bracis-19021	103	22	parameters	parameter	NOUN
bracis-19021	103	23	in	in	ADP
bracis-19021	103	24	maac(k	maac(k	NOUN
bracis-19021	103	25	)	)	PUNCT
bracis-19021	103	26	for	for	ADP
bracis-19021	103	27	\(k=3\	\(k=3\	NOUN
bracis-19021	103	28	)	)	PUNCT
bracis-19021	103	29	;	;	PUNCT
bracis-19021	103	30	dpg(k	dpg(k	PROPN
bracis-19021	103	31	)	)	PUNCT
bracis-19021	103	32	ignores	ignore	VERB
bracis-19021	103	33	these	these	DET
bracis-19021	103	34	dependencies	dependency	NOUN
bracis-19021	103	35	when	when	SCONJ
bracis-19021	103	36	backpropagating	backpropagate	VERB
bracis-19021	103	37	the	the	DET
bracis-19021	103	38	action	action	NOUN
bracis-19021	103	39	gradients	gradient	NOUN
bracis-19021	103	40	.	.	PUNCT
bracis-19021	104	1	full	full	ADJ
bracis-19021	104	2	size	size	NOUN
bracis-19021	104	3	image	image	NOUN
bracis-19021	104	4	4	4	NUM
bracis-19021	104	5	methodology	methodology	NOUN
bracis-19021	104	6	we	we	PRON
bracis-19021	104	7	now	now	ADV
bracis-19021	104	8	turn	turn	VERB
bracis-19021	104	9	to	to	ADP
bracis-19021	104	10	our	our	PRON
bracis-19021	104	11	research	research	NOUN
bracis-19021	104	12	goals	goal	NOUN
bracis-19021	104	13	and	and	CCONJ
bracis-19021	104	14	the	the	DET
bracis-19021	104	15	methods	method	NOUN
bracis-19021	104	16	employed	employ	VERB
bracis-19021	104	17	to	to	PART
bracis-19021	104	18	perform	perform	VERB
bracis-19021	104	19	the	the	DET
bracis-19021	104	20	proposed	propose	VERB
bracis-19021	104	21	empirical	empirical	ADJ
bracis-19021	104	22	analysis	analysis	NOUN
bracis-19021	104	23	.	.	PUNCT
bracis-19021	105	1	4.1	4.1	NUM
bracis-19021	105	2	scope	scope	NOUN
bracis-19021	105	3	of	of	ADP
bracis-19021	105	4	evaluation	evaluation	NOUN
bracis-19021	105	5	in	in	ADP
bracis-19021	105	6	this	this	DET
bracis-19021	105	7	work	work	NOUN
bracis-19021	105	8	,	,	PUNCT
bracis-19021	105	9	we	we	PRON
bracis-19021	105	10	propose	propose	VERB
bracis-19021	105	11	a	a	DET
bracis-19021	105	12	fine	fine	ADV
bracis-19021	105	13	-	-	PUNCT
bracis-19021	105	14	grained	grain	VERB
bracis-19021	105	15	analysis	analysis	NOUN
bracis-19021	105	16	of	of	ADP
bracis-19021	105	17	the	the	DET
bracis-19021	105	18	properties	property	NOUN
bracis-19021	105	19	of	of	ADP
bracis-19021	105	20	dpg(k	dpg(k	NOUN
bracis-19021	105	21	)	)	PUNCT
bracis-19021	105	22	and	and	CCONJ
bracis-19021	105	23	maac(k	maac(k	NOUN
bracis-19021	105	24	)	)	PUNCT
bracis-19021	105	25	in	in	ADP
bracis-19021	105	26	practice	practice	NOUN
bracis-19021	105	27	.	.	PUNCT
bracis-19021	106	1	we	we	PRON
bracis-19021	106	2	simplify	simplify	VERB
bracis-19021	106	3	our	our	PRON
bracis-19021	106	4	evaluation	evaluation	NOUN
bracis-19021	106	5	by	by	ADP
bracis-19021	106	6	using	use	VERB
bracis-19021	106	7	on	on	ADP
bracis-19021	106	8	-	-	PUNCT
bracis-19021	106	9	policy	policy	NOUN
bracis-19021	106	10	versions	version	NOUN
bracis-19021	106	11	of	of	ADP
bracis-19021	106	12	the	the	DET
bracis-19021	106	13	gradient	gradient	NOUN
bracis-19021	106	14	estimators	estimator	NOUN
bracis-19021	106	15	,	,	PUNCT
bracis-19021	106	16	i.e.	i.e.	X
bracis-19021	106	17	,	,	PUNCT
bracis-19021	106	18	by	by	ADP
bracis-19021	106	19	substituting	substitute	VERB
bracis-19021	106	20	\(d^{{\mu	\(d^{{\mu	NOUN
bracis-19021	106	21	}	}	PUNCT
bracis-19021	106	22	_	_	PUNCT
bracis-19021	106	23	{	{	PUNCT
bracis-19021	106	24	{	{	PUNCT
bracis-19021	106	25	\theta	\theta	PROPN
bracis-19021	106	26	}	}	PUNCT
bracis-19021	106	27	}	}	PUNCT
bracis-19021	106	28	}	}	PUNCT
bracis-19021	106	29	\	\	NOUN
bracis-19021	106	30	)	)	PUNCT
bracis-19021	106	31	for	for	ADP
bracis-19021	106	32	\(d^{\beta	\(d^{\beta	PROPN
bracis-19021	106	33	}	}	PUNCT
bracis-19021	106	34	\	\	NOUN
bracis-19021	106	35	)	)	PUNCT
bracis-19021	106	36	where	where	SCONJ
bracis-19021	106	37	present	present	ADJ
bracis-19021	106	38	.	.	PUNCT
bracis-19021	107	1	we	we	PRON
bracis-19021	107	2	also	also	ADV
bracis-19021	107	3	opted	opt	VERB
bracis-19021	107	4	for	for	ADP
bracis-19021	107	5	using	use	VERB
bracis-19021	107	6	perfect	perfect	ADJ
bracis-19021	107	7	models	model	NOUN
bracis-19021	107	8	of	of	ADP
bracis-19021	107	9	the	the	DET
bracis-19021	107	10	environment	environment	NOUN
bracis-19021	107	11	dynamics	dynamic	NOUN
bracis-19021	107	12	and	and	CCONJ
bracis-19021	107	13	rewards	reward	NOUN
bracis-19021	107	14	,	,	PUNCT
bracis-19021	107	15	instead	instead	ADV
bracis-19021	107	16	of	of	ADP
bracis-19021	107	17	learning	learn	VERB
bracis-19021	107	18	them	they	PRON
bracis-19021	107	19	from	from	ADP
bracis-19021	107	20	data	datum	NOUN
bracis-19021	107	21	,	,	PUNCT
bracis-19021	107	22	to	to	PART
bracis-19021	107	23	focus	focus	VERB
bracis-19021	107	24	on	on	ADP
bracis-19021	107	25	the	the	DET
bracis-19021	107	26	differences	difference	NOUN
bracis-19021	107	27	between	between	ADP
bracis-19021	107	28	gradient	gradient	ADJ
bracis-19021	107	29	estimators	estimator	NOUN
bracis-19021	107	30	.	.	PUNCT
bracis-19021	108	1	thus	thus	ADV
bracis-19021	108	2	,	,	PUNCT
bracis-19021	108	3	we	we	PRON
bracis-19021	108	4	approximate	approximate	VERB
bracis-19021	108	5	the	the	DET
bracis-19021	108	6	expectations	expectation	NOUN
bracis-19021	108	7	in	in	ADP
bracis-19021	108	8	eqs	eqs	PROPN
bracis-19021	108	9	.	.	PUNCT
bracis-19021	109	1	(	(	PUNCT
bracis-19021	109	2	6	6	NUM
bracis-19021	109	3	)	)	PUNCT
bracis-19021	109	4	and	and	CCONJ
bracis-19021	109	5	(	(	PUNCT
bracis-19021	109	6	7	7	X
bracis-19021	109	7	)	)	PUNCT
bracis-19021	109	8	via	via	ADP
bracis-19021	109	9	monte	monte	PROPN
bracis-19021	109	10	carlo	carlo	PROPN
bracis-19021	109	11	sampling	sampling	NOUN
bracis-19021	109	12	,	,	PUNCT
bracis-19021	109	13	using	use	VERB
bracis-19021	109	14	the	the	DET
bracis-19021	109	15	actual	actual	ADJ
bracis-19021	109	16	transition	transition	NOUN
bracis-19021	109	17	kernel	kernel	PROPN
bracis-19021	109	18	\(p^{*}\	\(p^{*}\	NOUN
bracis-19021	109	19	)	)	PUNCT
bracis-19021	109	20	and	and	CCONJ
bracis-19021	109	21	reward	reward	VERB
bracis-19021	109	22	function	function	NOUN
bracis-19021	109	23	\(r^{}\	\(r^{}\	NUM
bracis-19021	109	24	)	)	PUNCT
bracis-19021	109	25	,	,	PUNCT
bracis-19021	109	26	to	to	PART
bracis-19021	109	27	generate	generate	VERB
bracis-19021	109	28	(	(	PUNCT
bracis-19021	109	29	virtual	virtual	ADJ
bracis-19021	109	30	)	)	PUNCT
bracis-19021	109	31	transitions	transition	NOUN
bracis-19021	109	32	and	and	CCONJ
bracis-19021	109	33	compute	compute	VERB
bracis-19021	109	34	the	the	DET
bracis-19021	109	35	bootstrapped	bootstrappe	VERB
bracis-19021	109	36	returns	return	NOUN
bracis-19021	109	37	.	.	PUNCT
bracis-19021	110	1	one	one	PRON
bracis-19021	110	2	can	can	AUX
bracis-19021	110	3	view	view	VERB
bracis-19021	110	4	this	this	DET
bracis-19021	110	5	setting	setting	NOUN
bracis-19021	110	6	as	as	ADP
bracis-19021	110	7	the	the	DET
bracis-19021	110	8	best	good	ADJ
bracis-19021	110	9	possible	possible	ADJ
bracis-19021	110	10	case	case	NOUN
bracis-19021	110	11	in	in	ADP
bracis-19021	110	12	an	an	DET
bracis-19021	110	13	svg	svg	NOUN
bracis-19021	110	14	algorithm	algorithm	NOUN
bracis-19021	110	15	:	:	PUNCT
bracis-19021	110	16	when	when	SCONJ
bracis-19021	110	17	the	the	DET
bracis-19021	110	18	model	model	NOUN
bracis-19021	110	19	-	-	PUNCT
bracis-19021	110	20	learning	learn	VERB
bracis-19021	110	21	subroutine	subroutine	NOUN
bracis-19021	110	22	has	have	AUX
bracis-19021	110	23	perfectly	perfectly	ADV
bracis-19021	110	24	approximated	approximate	VERB
bracis-19021	110	25	the	the	DET
bracis-19021	110	26	true	true	ADJ
bracis-19021	110	27	mdp	mdp	PROPN
bracis-19021	110	28	,	,	PUNCT
bracis-19021	110	29	allowing	allow	VERB
bracis-19021	110	30	us	we	PRON
bracis-19021	110	31	to	to	PART
bracis-19021	110	32	focus	focus	VERB
bracis-19021	110	33	on	on	ADP
bracis-19021	110	34	the	the	DET
bracis-19021	110	35	gradient	gradient	NOUN
bracis-19021	110	36	estimation	estimation	NOUN
bracis-19021	110	37	analysis	analysis	NOUN
bracis-19021	110	38	.	.	PUNCT
bracis-19021	111	1	we	we	PRON
bracis-19021	111	2	also	also	ADV
bracis-19021	111	3	compute	compute	VERB
bracis-19021	111	4	the	the	DET
bracis-19021	111	5	true	true	ADJ
bracis-19021	111	6	action	action	NOUN
bracis-19021	111	7	-	-	PUNCT
bracis-19021	111	8	value	value	NOUN
bracis-19021	111	9	function	function	NOUN
bracis-19021	111	10	,	,	PUNCT
bracis-19021	111	11	required	require	VERB
bracis-19021	111	12	for	for	ADP
bracis-19021	111	13	the	the	DET
bracis-19021	111	14	k	k	ADJ
bracis-19021	111	15	-	-	PUNCT
bracis-19021	111	16	step	step	NOUN
bracis-19021	111	17	returns	return	NOUN
bracis-19021	111	18	in	in	ADP
bracis-19021	111	19	eqs	eqs	PROPN
bracis-19021	111	20	.	.	PUNCT
bracis-19021	112	1	(	(	PUNCT
bracis-19021	112	2	6	6	NUM
bracis-19021	112	3	)	)	PUNCT
bracis-19021	112	4	and	and	CCONJ
bracis-19021	112	5	(	(	PUNCT
bracis-19021	112	6	7	7	NUM
bracis-19021	112	7	)	)	PUNCT
bracis-19021	112	8	,	,	PUNCT
bracis-19021	112	9	recursively	recursively	ADV
bracis-19021	112	10	via	via	ADP
bracis-19021	112	11	dynamic	dynamic	ADJ
bracis-19021	112	12	programming	programming	NOUN
bracis-19021	112	13	(	(	PUNCT
bracis-19021	112	14	analogously	analogously	ADV
bracis-19021	112	15	to	to	ADP
bracis-19021	112	16	eq	eq	PROPN
bracis-19021	112	17	.	.	PUNCT
bracis-19021	113	1	(	(	PUNCT
bracis-19021	113	2	4	4	NUM
bracis-19021	113	3	)	)	PUNCT
bracis-19021	113	4	)	)	PUNCT
bracis-19021	113	5	.	.	PUNCT
bracis-19021	114	1	computing	compute	VERB
bracis-19021	114	2	the	the	DET
bracis-19021	114	3	ground	ground	NOUN
bracis-19021	114	4	-	-	PUNCT
bracis-19021	114	5	truth	truth	NOUN
bracis-19021	114	6	action	action	NOUN
bracis-19021	114	7	-	-	PUNCT
bracis-19021	114	8	value	value	NOUN
bracis-19021	114	9	function	function	NOUN
bracis-19021	114	10	allows	allow	VERB
bracis-19021	114	11	us	we	PRON
bracis-19021	114	12	to	to	PART
bracis-19021	114	13	further	far	ADV
bracis-19021	114	14	isolate	isolate	VERB
bracis-19021	114	15	any	any	DET
bracis-19021	114	16	observed	observe	VERB
bracis-19021	114	17	differences	difference	NOUN
bracis-19021	114	18	between	between	ADP
bracis-19021	114	19	the	the	DET
bracis-19021	114	20	estimators	estimator	NOUN
bracis-19021	114	21	as	as	ADP
bracis-19021	114	22	a	a	DET
bracis-19021	114	23	consequence	consequence	NOUN
bracis-19021	114	24	of	of	ADP
bracis-19021	114	25	their	their	PRON
bracis-19021	114	26	properties	property	NOUN
bracis-19021	114	27	alone	alone	ADV
bracis-19021	114	28	.	.	PUNCT
bracis-19021	115	1	4.2	4.2	NUM
bracis-19021	115	2	randomized	randomize	VERB
bracis-19021	115	3	environments	environment	NOUN
bracis-19021	115	4	and	and	CCONJ
bracis-19021	115	5	policies	policy	NOUN
bracis-19021	115	6	to	to	PART
bracis-19021	115	7	test	test	VERB
bracis-19021	115	8	gradient	gradient	NOUN
bracis-19021	115	9	estimation	estimation	NOUN
bracis-19021	115	10	across	across	ADP
bracis-19021	115	11	a	a	DET
bracis-19021	115	12	wide	wide	ADJ
bracis-19021	115	13	variety	variety	NOUN
bracis-19021	115	14	of	of	ADP
bracis-19021	115	15	scenarios	scenario	NOUN
bracis-19021	115	16	,	,	PUNCT
bracis-19021	115	17	we	we	PRON
bracis-19021	115	18	define	define	VERB
bracis-19021	115	19	how	how	SCONJ
bracis-19021	115	20	to	to	PART
bracis-19021	115	21	sample	sample	VERB
bracis-19021	115	22	lqg	lqg	NOUN
bracis-19021	115	23	instances	instance	NOUN
bracis-19021	115	24	,	,	PUNCT
bracis-19021	115	25	\({\mathcal	\({\mathcal	PROPN
bracis-19021	115	26	m}=	m}=	PROPN
bracis-19021	115	27	(	(	PUNCT
bracis-19021	115	28	{	{	PUNCT
bracis-19021	115	29	\mathcal	\mathcal	PROPN
bracis-19021	115	30	s	s	PART
bracis-19021	115	31	}	}	PUNCT
bracis-19021	115	32	,	,	PUNCT
bracis-19021	115	33	{	{	PUNCT
bracis-19021	115	34	\mathcal	\mathcal	NOUN
bracis-19021	115	35	a	a	X
bracis-19021	115	36	}	}	PUNCT
bracis-19021	115	37	,	,	PUNCT
bracis-19021	115	38	r^	r^	NOUN
bracis-19021	115	39	{	{	PUNCT
bracis-19021	115	40	}	}	PUNCT
bracis-19021	115	41	,	,	PUNCT
bracis-19021	115	42	p^	p^	PROPN
bracis-19021	115	43	{	{	PUNCT
bracis-19021	115	44	*	*	PUNCT
bracis-19021	115	45	}	}	PUNCT
bracis-19021	115	46	,	,	PUNCT
bracis-19021	115	47	{	{	PUNCT
bracis-19021	115	48	\rho	\rho	NOUN
bracis-19021	115	49	}	}	PUNCT
bracis-19021	115	50	)	)	PUNCT
bracis-19021	115	51	\	\	PROPN
bracis-19021	115	52	)	)	PUNCT
bracis-19021	115	53	,	,	PUNCT
bracis-19021	115	54	to	to	PART
bracis-19021	115	55	run	run	VERB
bracis-19021	115	56	our	our	PRON
bracis-19021	115	57	experiments	experiment	NOUN
bracis-19021	115	58	on	on	ADP
bracis-19021	115	59	.	.	PUNCT
bracis-19021	116	1	the	the	DET
bracis-19021	116	2	main	main	ADJ
bracis-19021	116	3	configurations	configuration	NOUN
bracis-19021	116	4	for	for	ADP
bracis-19021	116	5	our	our	PRON
bracis-19021	116	6	procedure	procedure	NOUN
bracis-19021	116	7	are	be	AUX
bracis-19021	116	8	:	:	PUNCT
bracis-19021	116	9	state	state	NOUN
bracis-19021	116	10	dimension	dimension	NOUN
bracis-19021	116	11	(	(	PUNCT
bracis-19021	116	12	\(n\	\(n\	PROPN
bracis-19021	116	13	)	)	PUNCT
bracis-19021	116	14	)	)	PUNCT
bracis-19021	116	15	,	,	PUNCT
bracis-19021	116	16	action	action	NOUN
bracis-19021	116	17	dimension	dimension	NOUN
bracis-19021	116	18	(	(	PUNCT
bracis-19021	116	19	\(d\	\(d\	NOUN
bracis-19021	116	20	)	)	PUNCT
bracis-19021	116	21	)	)	PUNCT
bracis-19021	116	22	and	and	CCONJ
bracis-19021	116	23	time	time	NOUN
bracis-19021	116	24	horizon	horizon	PROPN
bracis-19021	116	25	(	(	PUNCT
bracis-19021	116	26	\(h\	\(h\	NOUN
bracis-19021	116	27	)	)	PUNCT
bracis-19021	116	28	)	)	PUNCT
bracis-19021	116	29	.	.	PUNCT
bracis-19021	117	1	from	from	ADP
bracis-19021	117	2	these	these	DET
bracis-19021	117	3	parameters	parameter	NOUN
bracis-19021	117	4	we	we	PRON
bracis-19021	117	5	define	define	VERB
bracis-19021	117	6	the	the	DET
bracis-19021	117	7	state	state	NOUN
bracis-19021	117	8	space	space	NOUN
bracis-19021	117	9	\({\mathcal	\({\mathcal	PROPN
bracis-19021	117	10	s}={\mathbb	s}={\mathbb	NOUN
bracis-19021	117	11	r}^n\	r}^n\	NOUN
bracis-19021	117	12	)	)	PUNCT
bracis-19021	117	13	,	,	PUNCT
bracis-19021	117	14	action	action	NOUN
bracis-19021	117	15	space	space	NOUN
bracis-19021	117	16	\({\mathcal	\({\mathcal	PROPN
bracis-19021	117	17	a}={\mathbb	a}={\mathbb	NOUN
bracis-19021	117	18	r}^d\	r}^d\	NUM
bracis-19021	117	19	)	)	PUNCT
bracis-19021	117	20	and	and	CCONJ
bracis-19021	117	21	timesteps	timestep	NOUN
bracis-19021	117	22	.	.	PUNCT
bracis-19021	118	1	the	the	DET
bracis-19021	118	2	transition	transition	NOUN
bracis-19021	118	3	dynamics	dynamic	NOUN
bracis-19021	118	4	are	be	AUX
bracis-19021	118	5	stationary	stationary	ADJ
bracis-19021	118	6	,	,	PUNCT
bracis-19021	118	7	sharing	share	VERB
bracis-19021	118	8	\(\mathbf	\(\mathbf	NOUN
bracis-19021	118	9	{	{	PUNCT
bracis-19021	118	10	f	f	NOUN
bracis-19021	118	11	}	}	PUNCT
bracis-19021	118	12	,	,	PUNCT
bracis-19021	118	13	\mathbf	\mathbf	PROPN
bracis-19021	118	14	{	{	PUNCT
bracis-19021	118	15	f	f	X
bracis-19021	118	16	}	}	PUNCT
bracis-19021	118	17	,	,	PUNCT
bracis-19021	118	18	\mathbf	\mathbf	PROPN
bracis-19021	118	19	{	{	PUNCT
bracis-19021	118	20	\sigma	\sigma	PROPN
bracis-19021	118	21	}	}	PUNCT
bracis-19021	118	22	\	\	NOUN
bracis-19021	118	23	)	)	PUNCT
bracis-19021	118	24	across	across	ADP
bracis-19021	118	25	all	all	DET
bracis-19021	118	26	timesteps	timestep	NOUN
bracis-19021	118	27	.	.	PUNCT
bracis-19021	119	1	the	the	DET
bracis-19021	119	2	coefficients	coefficient	NOUN
bracis-19021	119	3	\(\mathbf	\(\mathbf	VERB
bracis-19021	119	4	{	{	PUNCT
bracis-19021	119	5	f_s}\	f_s}\	X
bracis-19021	119	6	)	)	PUNCT
bracis-19021	119	7	and	and	CCONJ
bracis-19021	119	8	\(\mathbf	\(\mathbf	VERB
bracis-19021	119	9	{	{	PUNCT
bracis-19021	119	10	f_a}\	f_a}\	X
bracis-19021	119	11	)	)	PUNCT
bracis-19021	119	12	are	be	AUX
bracis-19021	119	13	initialized	initialize	VERB
bracis-19021	119	14	so	so	SCONJ
bracis-19021	119	15	that	that	SCONJ
bracis-19021	119	16	the	the	DET
bracis-19021	119	17	system	system	NOUN
bracis-19021	119	18	may	may	AUX
bracis-19021	119	19	be	be	AUX
bracis-19021	119	20	unstable	unstable	ADJ
bracis-19021	119	21	,	,	PUNCT
bracis-19021	119	22	i.e.	i.e.	X
bracis-19021	119	23	,	,	PUNCT
bracis-19021	119	24	with	with	ADP
bracis-19021	119	25	some	some	DET
bracis-19021	119	26	eigenvalues	eigenvalue	NOUN
bracis-19021	119	27	of	of	ADP
bracis-19021	119	28	\(\mathbf	\(\mathbf	PROPN
bracis-19021	119	29	{	{	PUNCT
bracis-19021	119	30	f_s}\	f_s}\	X
bracis-19021	119	31	)	)	PUNCT
bracis-19021	119	32	having	have	VERB
bracis-19021	119	33	magnitude	magnitude	NOUN
bracis-19021	119	34	greater	great	ADJ
bracis-19021	119	35	or	or	CCONJ
bracis-19021	119	36	equal	equal	ADJ
bracis-19021	119	37	to	to	ADP
bracis-19021	119	38	1	1	NUM
bracis-19021	119	39	,	,	PUNCT
bracis-19021	119	40	but	but	CCONJ
bracis-19021	119	41	always	always	ADV
bracis-19021	119	42	controllable	controllable	ADJ
bracis-19021	119	43	,	,	PUNCT
bracis-19021	119	44	meaning	mean	VERB
bracis-19021	119	45	there	there	PRON
bracis-19021	119	46	is	be	VERB
bracis-19021	119	47	a	a	DET
bracis-19021	119	48	dynamic	dynamic	ADJ
bracis-19021	119	49	gain	gain	NOUN
bracis-19021	119	50	\(\mathbf	\(\mathbf	VERB
bracis-19021	119	51	{	{	PUNCT
bracis-19021	119	52	k}\	k}\	NOUN
bracis-19021	119	53	)	)	PUNCT
bracis-19021	119	54	such	such	ADJ
bracis-19021	119	55	that	that	SCONJ
bracis-19021	119	56	the	the	DET
bracis-19021	119	57	eigenvalues	eigenvalue	NOUN
bracis-19021	119	58	of	of	ADP
bracis-19021	119	59	\((\mathbf	\((\mathbf	PROPN
bracis-19021	119	60	{	{	PUNCT
bracis-19021	119	61	f_s}+	f_s}+	X
bracis-19021	119	62	\mathbf	\mathbf	PROPN
bracis-19021	119	63	{	{	PUNCT
bracis-19021	119	64	f_a}\mathbf	f_a}\mathbf	PROPN
bracis-19021	119	65	{	{	PUNCT
bracis-19021	119	66	k})\	k})\	NOUN
bracis-19021	119	67	)	)	PUNCT
bracis-19021	119	68	have	have	AUX
bracis-19021	119	69	magnitude	magnitude	NOUN
bracis-19021	119	70	less	less	ADJ
bracis-19021	119	71	than	than	ADP
bracis-19021	119	72	1	1	NUM
bracis-19021	119	73	.	.	PUNCT
bracis-19021	120	1	this	this	PRON
bracis-19021	120	2	ensures	ensure	VERB
bracis-19021	120	3	we	we	PRON
bracis-19021	120	4	are	be	AUX
bracis-19021	120	5	able	able	ADJ
bracis-19021	120	6	to	to	PART
bracis-19021	120	7	emulate	emulate	VERB
bracis-19021	120	8	real	real	ADJ
bracis-19021	120	9	-	-	PUNCT
bracis-19021	120	10	world	world	NOUN
bracis-19021	120	11	scenarios	scenario	NOUN
bracis-19021	120	12	where	where	SCONJ
bracis-19021	120	13	uncontrolled	uncontrolled	ADJ
bracis-19021	120	14	state	state	NOUN
bracis-19021	120	15	variables	variable	NOUN
bracis-19021	120	16	and	and	CCONJ
bracis-19021	120	17	costs	cost	NOUN
bracis-19021	120	18	may	may	AUX
bracis-19021	120	19	diverge	diverge	VERB
bracis-19021	120	20	to	to	PART
bracis-19021	120	21	infinity	infinity	VERB
bracis-19021	120	22	,	,	PUNCT
bracis-19021	120	23	while	while	SCONJ
bracis-19021	120	24	ensuring	ensure	VERB
bracis-19021	120	25	there	there	PRON
bracis-19021	120	26	exists	exist	VERB
bracis-19021	120	27	a	a	DET
bracis-19021	120	28	policy	policy	NOUN
bracis-19021	120	29	which	which	PRON
bracis-19021	120	30	can	can	AUX
bracis-19021	120	31	stabilize	stabilize	VERB
bracis-19021	120	32	the	the	DET
bracis-19021	120	33	system	system	NOUN
bracis-19021	120	34	[	[	X
bracis-19021	120	35	21	21	NUM
bracis-19021	120	36	]	]	PUNCT
bracis-19021	120	37	.	.	PUNCT
bracis-19021	121	1	finally	finally	ADV
bracis-19021	121	2	,	,	PUNCT
bracis-19021	121	3	we	we	PRON
bracis-19021	121	4	fix	fix	VERB
bracis-19021	121	5	the	the	DET
bracis-19021	121	6	transition	transition	NOUN
bracis-19021	121	7	bias	bias	NOUN
bracis-19021	121	8	to	to	PART
bracis-19021	121	9	\(\mathbf	\(\mathbf	VERB
bracis-19021	121	10	{	{	PUNCT
bracis-19021	121	11	f}=	f}=	NOUN
bracis-19021	121	12	{	{	PUNCT
bracis-19021	121	13	\mathbf	\mathbf	PROPN
bracis-19021	121	14	{	{	PUNCT
bracis-19021	121	15	0}}\	0}}\	NUM
bracis-19021	121	16	)	)	PUNCT
bracis-19021	121	17	and	and	CCONJ
bracis-19021	121	18	the	the	DET
bracis-19021	121	19	gaussian	gaussian	ADJ
bracis-19021	121	20	covariance	covariance	NOUN
bracis-19021	121	21	to	to	ADP
bracis-19021	121	22	the	the	DET
bracis-19021	121	23	identity	identity	NOUN
bracis-19021	121	24	matrix	matrix	NOUN
bracis-19021	121	25	,	,	PUNCT
bracis-19021	121	26	\(\mathbf	\(\mathbf	VERB
bracis-19021	121	27	{	{	PUNCT
bracis-19021	121	28	\sigma	\sigma	PROPN
bracis-19021	121	29	}	}	PUNCT
bracis-19021	121	30	=	=	SYM
bracis-19021	121	31	{	{	PUNCT
bracis-19021	121	32	i}\	i}\	NUM
bracis-19021	121	33	)	)	PUNCT
bracis-19021	121	34	.	.	PUNCT
bracis-19021	122	1	the	the	DET
bracis-19021	122	2	initial	initial	ADJ
bracis-19021	122	3	state	state	NOUN
bracis-19021	122	4	distribution	distribution	NOUN
bracis-19021	122	5	is	be	AUX
bracis-19021	122	6	always	always	ADV
bracis-19021	122	7	initialized	initialize	VERB
bracis-19021	122	8	as	as	ADP
bracis-19021	122	9	a	a	DET
bracis-19021	122	10	standard	standard	ADJ
bracis-19021	122	11	gaussian	gaussian	ADJ
bracis-19021	122	12	distribution	distribution	NOUN
bracis-19021	122	13	:	:	PUNCT
bracis-19021	122	14	.	.	PUNCT
bracis-19021	123	1	as	as	ADP
bracis-19021	123	2	for	for	ADP
bracis-19021	123	3	the	the	DET
bracis-19021	123	4	reward	reward	NOUN
bracis-19021	123	5	parameters	parameter	NOUN
bracis-19021	123	6	,	,	PUNCT
bracis-19021	123	7	we	we	PRON
bracis-19021	123	8	initialize	initialize	VERB
bracis-19021	123	9	both	both	DET
bracis-19021	123	10	\({{\mathbf	\({{\mathbf	PROPN
bracis-19021	123	11	{	{	PUNCT
bracis-19021	123	12	c}}_{{\mathbf	c}}_{{\mathbf	PROPN
bracis-19021	123	13	{	{	PUNCT
bracis-19021	123	14	s}}{\mathbf	s}}{\mathbf	PROPN
bracis-19021	123	15	{	{	PUNCT
bracis-19021	123	16	s}}}}\	s}}}}\	PROPN
bracis-19021	123	17	)	)	PUNCT
bracis-19021	123	18	and	and	CCONJ
bracis-19021	123	19	\({{\mathbf	\({{\mathbf	PROPN
bracis-19021	123	20	{	{	PUNCT
bracis-19021	123	21	c}}_{{\mathbf	c}}_{{\mathbf	PROPN
bracis-19021	123	22	{	{	PUNCT
bracis-19021	123	23	a}}{\mathbf	a}}{\mathbf	NOUN
bracis-19021	123	24	{	{	PUNCT
bracis-19021	123	25	a}}}}\	a}}}}\	PROPN
bracis-19021	123	26	)	)	PUNCT
bracis-19021	123	27	(	(	PUNCT
bracis-19021	123	28	see	see	VERB
bracis-19021	123	29	eq	eq	ADP
bracis-19021	123	30	.	.	PUNCT
bracis-19021	124	1	(	(	PUNCT
bracis-19021	124	2	2	2	NUM
bracis-19021	124	3	)	)	PUNCT
bracis-19021	124	4	)	)	PUNCT
bracis-19021	124	5	as	as	ADP
bracis-19021	124	6	random	random	ADJ
bracis-19021	124	7	symmetric	symmetric	ADJ
bracis-19021	124	8	positive	positive	ADJ
bracis-19021	124	9	definite	definite	ADJ
bracis-19021	124	10	matrices	matrix	NOUN
bracis-19021	124	11	,	,	PUNCT
bracis-19021	124	12	sampled	sample	VERB
bracis-19021	124	13	via	via	ADP
bracis-19021	124	14	the	the	DET
bracis-19021	124	15	scikit-​learn	scikit-​learn	PRON
bracis-19021	124	16	library	library	NOUN
bracis-19021	124	17	for	for	ADP
bracis-19021	124	18	machine	machine	NOUN
bracis-19021	124	19	learning	learning	NOUN
bracis-19021	124	20	in	in	ADP
bracis-19021	124	21	python	python	NOUN
bracis-19021	125	1	[	[	X
bracis-19021	125	2	19].footnote	19].footnote	PROPN
bracis-19021	125	3	2	2	NUM
bracis-19021	125	4	since	since	SCONJ
bracis-19021	125	5	we	we	PRON
bracis-19021	125	6	consider	consider	VERB
bracis-19021	125	7	the	the	DET
bracis-19021	125	8	problem	problem	NOUN
bracis-19021	125	9	of	of	ADP
bracis-19021	125	10	estimating	estimate	VERB
bracis-19021	125	11	value	value	NOUN
bracis-19021	125	12	gradients	gradient	NOUN
bracis-19021	125	13	for	for	ADP
bracis-19021	125	14	linear	linear	ADJ
bracis-19021	125	15	policies	policy	NOUN
bracis-19021	125	16	in	in	ADP
bracis-19021	125	17	lqgs	lqg	NOUN
bracis-19021	125	18	,	,	PUNCT
bracis-19021	125	19	we	we	PRON
bracis-19021	125	20	also	also	ADV
bracis-19021	125	21	define	define	VERB
bracis-19021	125	22	a	a	DET
bracis-19021	125	23	procedure	procedure	NOUN
bracis-19021	125	24	to	to	PART
bracis-19021	125	25	generate	generate	VERB
bracis-19021	125	26	randomized	randomized	ADJ
bracis-19021	125	27	policies	policy	NOUN
bracis-19021	125	28	.	.	PUNCT
bracis-19021	126	1	we	we	PRON
bracis-19021	126	2	start	start	VERB
bracis-19021	126	3	by	by	ADP
bracis-19021	126	4	initializing	initialize	VERB
bracis-19021	126	5	all	all	DET
bracis-19021	126	6	dynamic	dynamic	ADJ
bracis-19021	126	7	gains	gain	NOUN
bracis-19021	126	8	\(\mathbf	\(\mathbf	VERB
bracis-19021	126	9	{	{	PUNCT
bracis-19021	126	10	k}_t	k}_t	NOUN
bracis-19021	126	11	=	=	SYM
bracis-19021	126	12	\mathbf	\mathbf	PROPN
bracis-19021	126	13	{	{	PUNCT
bracis-19021	126	14	k}\	k}\	NOUN
bracis-19021	126	15	)	)	PUNCT
bracis-19021	126	16	so	so	SCONJ
bracis-19021	126	17	that	that	SCONJ
bracis-19021	126	18	\(\mathbf	\(\mathbf	VERB
bracis-19021	126	19	{	{	PUNCT
bracis-19021	126	20	k}\	k}\	NOUN
bracis-19021	126	21	)	)	PUNCT
bracis-19021	126	22	stabilizes	stabilize	VERB
bracis-19021	126	23	the	the	DET
bracis-19021	126	24	system	system	NOUN
bracis-19021	126	25	.	.	PUNCT
bracis-19021	127	1	this	this	PRON
bracis-19021	127	2	is	be	AUX
bracis-19021	127	3	done	do	VERB
bracis-19021	127	4	by	by	ADP
bracis-19021	127	5	first	first	ADJ
bracis-19021	127	6	sampling	sample	VERB
bracis-19021	127	7	target	target	NOUN
bracis-19021	127	8	eigenvalues	eigenvalue	VERB
bracis-19021	127	9	uniformly	uniformly	ADV
bracis-19021	127	10	in	in	ADP
bracis-19021	127	11	the	the	DET
bracis-19021	127	12	interval	interval	NOUN
bracis-19021	127	13	(	(	PUNCT
bracis-19021	127	14	0	0	NUM
bracis-19021	127	15	,	,	PUNCT
bracis-19021	127	16	 	 	SPACE
bracis-19021	127	17	1	1	NUM
bracis-19021	127	18	)	)	PUNCT
bracis-19021	127	19	and	and	CCONJ
bracis-19021	127	20	then	then	ADV
bracis-19021	127	21	using	use	VERB
bracis-19021	127	22	the	the	DET
bracis-19021	127	23	scipy	scipy	NOUN
bracis-19021	127	24	library	library	NOUN
bracis-19021	127	25	to	to	PART
bracis-19021	127	26	compute	compute	VERB
bracis-19021	127	27	\(\mathbf	\(\mathbf	VERB
bracis-19021	127	28	{	{	PUNCT
bracis-19021	127	29	k}\	k}\	NOUN
bracis-19021	127	30	)	)	PUNCT
bracis-19021	127	31	that	that	PRON
bracis-19021	127	32	places	place	VERB
bracis-19021	127	33	the	the	DET
bracis-19021	127	34	eigenvalues	eigenvalue	NOUN
bracis-19021	127	35	of	of	ADP
bracis-19021	127	36	\((\mathbf	\((\mathbf	PROPN
bracis-19021	127	37	{	{	PUNCT
bracis-19021	127	38	f_s}+	f_s}+	X
bracis-19021	127	39	\mathbf	\mathbf	PROPN
bracis-19021	127	40	{	{	PUNCT
bracis-19021	127	41	f_a}\mathbf	f_a}\mathbf	PROPN
bracis-19021	127	42	{	{	PUNCT
bracis-19021	127	43	k})\	k})\	NOUN
bracis-19021	127	44	)	)	PUNCT
bracis-19021	127	45	in	in	ADP
bracis-19021	127	46	the	the	DET
bracis-19021	127	47	desired	desire	VERB
bracis-19021	127	48	targets	target	NOUN
bracis-19021	127	49	[	[	X
bracis-19021	127	50	26	26	NUM
bracis-19021	127	51	,	,	PUNCT
bracis-19021	127	52	29].footnote	29].footnote	PROPN
bracis-19021	127	53	3	3	NUM
bracis-19021	127	54	this	this	DET
bracis-19021	127	55	process	process	NOUN
bracis-19021	127	56	ensures	ensure	VERB
bracis-19021	127	57	the	the	DET
bracis-19021	127	58	resulting	result	VERB
bracis-19021	127	59	policy	policy	NOUN
bracis-19021	127	60	is	be	AUX
bracis-19021	127	61	safe	safe	ADJ
bracis-19021	127	62	to	to	PART
bracis-19021	127	63	collect	collect	VERB
bracis-19021	127	64	data	datum	NOUN
bracis-19021	127	65	in	in	ADP
bracis-19021	127	66	the	the	DET
bracis-19021	127	67	environment	environment	NOUN
bracis-19021	127	68	without	without	ADP
bracis-19021	127	69	having	have	VERB
bracis-19021	127	70	state	state	NOUN
bracis-19021	127	71	variables	variable	NOUN
bracis-19021	127	72	and	and	CCONJ
bracis-19021	127	73	costs	cost	VERB
bracis-19021	127	74	diverge	diverge	VERB
bracis-19021	127	75	to	to	PART
bracis-19021	127	76	infinity	infinity	NOUN
bracis-19021	127	77	.	.	PUNCT
bracis-19021	128	1	the	the	DET
bracis-19021	128	2	generating	generate	VERB
bracis-19021	128	3	procedure	procedure	NOUN
bracis-19021	128	4	also	also	ADV
bracis-19021	128	5	serves	serve	VERB
bracis-19021	128	6	to	to	PART
bracis-19021	128	7	mimic	mimic	VERB
bracis-19021	128	8	practical	practical	ADJ
bracis-19021	128	9	situations	situation	NOUN
bracis-19021	128	10	where	where	SCONJ
bracis-19021	128	11	engineers	engineer	NOUN
bracis-19021	128	12	devise	devise	VERB
bracis-19021	128	13	a	a	DET
bracis-19021	128	14	policy	policy	NOUN
bracis-19021	128	15	which	which	PRON
bracis-19021	128	16	can	can	AUX
bracis-19021	128	17	keep	keep	VERB
bracis-19021	128	18	a	a	DET
bracis-19021	128	19	system	system	NOUN
bracis-19021	128	20	stable	stable	ADJ
bracis-19021	128	21	,	,	PUNCT
bracis-19021	128	22	but	but	CCONJ
bracis-19021	128	23	is	be	AUX
bracis-19021	128	24	not	not	PART
bracis-19021	128	25	able	able	ADJ
bracis-19021	128	26	to	to	PART
bracis-19021	128	27	optimize	optimize	VERB
bracis-19021	128	28	running	running	NOUN
bracis-19021	128	29	costs	cost	NOUN
bracis-19021	128	30	,	,	PUNCT
bracis-19021	128	31	which	which	PRON
bracis-19021	128	32	is	be	AUX
bracis-19021	128	33	where	where	SCONJ
bracis-19021	128	34	rl	rl	PROPN
bracis-19021	128	35	can	can	AUX
bracis-19021	128	36	serve	serve	VERB
bracis-19021	128	37	to	to	ADP
bracis-19021	128	38	fine	fine	ADJ
bracis-19021	128	39	-	-	PUNCT
bracis-19021	128	40	tune	tune	NOUN
bracis-19021	128	41	it	it	PRON
bracis-19021	128	42	.	.	PUNCT
bracis-19021	129	1	finally	finally	ADV
bracis-19021	129	2	,	,	PUNCT
bracis-19021	129	3	we	we	PRON
bracis-19021	129	4	initialize	initialize	VERB
bracis-19021	129	5	all	all	DET
bracis-19021	129	6	static	static	ADJ
bracis-19021	129	7	gains	gain	NOUN
bracis-19021	129	8	as	as	SCONJ
bracis-19021	129	9	\(\mathbf	\(\mathbf	VERB
bracis-19021	129	10	{	{	PUNCT
bracis-19021	129	11	k}_t	k}_t	NOUN
bracis-19021	129	12	=	=	SYM
bracis-19021	129	13	{	{	PUNCT
bracis-19021	129	14	\mathbf	\mathbf	PROPN
bracis-19021	129	15	{	{	PUNCT
bracis-19021	129	16	0}}\	0}}\	NUM
bracis-19021	129	17	)	)	PUNCT
bracis-19021	129	18	.	.	PUNCT
bracis-19021	130	1	5	5	NUM
bracis-19021	130	2	empirical	empirical	ADJ
bracis-19021	130	3	analysis	analysis	NOUN
bracis-19021	130	4	we	we	PRON
bracis-19021	130	5	analyze	analyze	VERB
bracis-19021	130	6	the	the	DET
bracis-19021	130	7	behavior	behavior	NOUN
bracis-19021	130	8	of	of	ADP
bracis-19021	130	9	each	each	DET
bracis-19021	130	10	estimator	estimator	NOUN
bracis-19021	130	11	on	on	ADP
bracis-19021	130	12	two	two	NUM
bracis-19021	130	13	main	main	ADJ
bracis-19021	130	14	settings	setting	NOUN
bracis-19021	130	15	:	:	PUNCT
bracis-19021	130	16	(	(	PUNCT
bracis-19021	130	17	i	i	NOUN
bracis-19021	130	18	)	)	PUNCT
bracis-19021	130	19	gradient	gradient	NOUN
bracis-19021	130	20	estimation	estimation	NOUN
bracis-19021	130	21	for	for	ADP
bracis-19021	130	22	fixed	fix	VERB
bracis-19021	130	23	policies	policy	NOUN
bracis-19021	130	24	and	and	CCONJ
bracis-19021	130	25	(	(	PUNCT
bracis-19021	130	26	ii	ii	NOUN
bracis-19021	130	27	)	)	PUNCT
bracis-19021	130	28	impact	impact	NOUN
bracis-19021	130	29	of	of	ADP
bracis-19021	130	30	gradient	gradient	ADJ
bracis-19021	130	31	quality	quality	NOUN
bracis-19021	130	32	on	on	ADP
bracis-19021	130	33	policy	policy	NOUN
bracis-19021	130	34	optimization	optimization	NOUN
bracis-19021	130	35	.	.	PUNCT
bracis-19021	131	1	5.1	5.1	NUM
bracis-19021	131	2	gradient	gradient	NOUN
bracis-19021	131	3	estimation	estimation	NOUN
bracis-19021	131	4	for	for	ADP
bracis-19021	131	5	fixed	fix	VERB
bracis-19021	131	6	policies	policy	NOUN
bracis-19021	131	7	following	follow	VERB
bracis-19021	131	8	previous	previous	ADJ
bracis-19021	131	9	work	work	NOUN
bracis-19021	131	10	on	on	ADP
bracis-19021	131	11	model	model	ADJ
bracis-19021	131	12	-	-	PUNCT
bracis-19021	131	13	free	free	ADJ
bracis-19021	131	14	policy	policy	NOUN
bracis-19021	131	15	gradients	gradient	NOUN
bracis-19021	131	16	[	[	X
bracis-19021	131	17	11	11	NUM
bracis-19021	131	18	]	]	PUNCT
bracis-19021	131	19	,	,	PUNCT
bracis-19021	131	20	we	we	PRON
bracis-19021	131	21	evaluate	evaluate	VERB
bracis-19021	131	22	the	the	DET
bracis-19021	131	23	quality	quality	NOUN
bracis-19021	131	24	of	of	ADP
bracis-19021	131	25	the	the	DET
bracis-19021	131	26	gradient	gradient	NOUN
bracis-19021	131	27	estimates	estimate	NOUN
bracis-19021	131	28	,	,	PUNCT
bracis-19021	131	29	for	for	ADP
bracis-19021	131	30	a	a	DET
bracis-19021	131	31	given	give	VERB
bracis-19021	131	32	policy	policy	NOUN
bracis-19021	131	33	,	,	PUNCT
bracis-19021	131	34	produced	produce	VERB
bracis-19021	131	35	by	by	ADP
bracis-19021	131	36	each	each	DET
bracis-19021	131	37	estimator	estimator	NOUN
bracis-19021	131	38	using	use	VERB
bracis-19021	131	39	two	two	NUM
bracis-19021	131	40	metrics	metric	NOUN
bracis-19021	131	41	:	:	PUNCT
bracis-19021	131	42	(	(	PUNCT
bracis-19021	131	43	i	i	NOUN
bracis-19021	131	44	)	)	PUNCT
bracis-19021	131	45	the	the	DET
bracis-19021	131	46	average	average	ADJ
bracis-19021	131	47	cosine	cosine	NOUN
bracis-19021	131	48	similarity	similarity	NOUN
bracis-19021	131	49	with	with	ADP
bracis-19021	131	50	the	the	DET
bracis-19021	131	51	true	true	ADJ
bracis-19021	131	52	policy	policy	NOUN
bracis-19021	131	53	gradient	gradient	NOUN
bracis-19021	131	54	and	and	CCONJ
bracis-19021	131	55	(	(	PUNCT
bracis-19021	131	56	ii	ii	NOUN
bracis-19021	131	57	)	)	PUNCT
bracis-19021	131	58	the	the	DET
bracis-19021	131	59	average	average	ADJ
bracis-19021	131	60	pairwise	pairwise	PROPN
bracis-19021	131	61	cosine	cosine	NOUN
bracis-19021	131	62	similarity	similarity	NOUN
bracis-19021	131	63	.	.	PUNCT
bracis-19021	132	1	the	the	DET
bracis-19021	132	2	first	first	ADJ
bracis-19021	132	3	metric	metric	NOUN
bracis-19021	132	4	is	be	AUX
bracis-19021	132	5	a	a	DET
bracis-19021	132	6	measure	measure	NOUN
bracis-19021	132	7	of	of	ADP
bracis-19021	132	8	gradient	gradient	ADJ
bracis-19021	132	9	accuracy	accuracy	NOUN
bracis-19021	132	10	and	and	CCONJ
bracis-19021	132	11	we	we	PRON
bracis-19021	132	12	denote	denote	VERB
bracis-19021	132	13	it	it	PRON
bracis-19021	132	14	as	as	ADP
bracis-19021	132	15	such	such	ADJ
bracis-19021	132	16	in	in	ADP
bracis-19021	132	17	the	the	DET
bracis-19021	132	18	following	follow	VERB
bracis-19021	132	19	plots	plot	NOUN
bracis-19021	132	20	.	.	PUNCT
bracis-19021	133	1	for	for	ADP
bracis-19021	133	2	a	a	DET
bracis-19021	133	3	given	give	VERB
bracis-19021	133	4	minibatch	minibatch	NOUN
bracis-19021	133	5	size	size	NOUN
bracis-19021	133	6	b	b	NOUN
bracis-19021	133	7	and	and	CCONJ
bracis-19021	133	8	step	step	NOUN
bracis-19021	133	9	size	size	NOUN
bracis-19021	133	10	k	k	PROPN
bracis-19021	133	11	,	,	PUNCT
bracis-19021	133	12	we	we	PRON
bracis-19021	133	13	compute	compute	VERB
bracis-19021	133	14	10	10	NUM
bracis-19021	133	15	estimates	estimate	NOUN
bracis-19021	133	16	of	of	ADP
bracis-19021	133	17	the	the	DET
bracis-19021	133	18	gradient	gradient	NOUN
bracis-19021	133	19	,	,	PUNCT
bracis-19021	134	1	each	each	DET
bracis-19021	134	2	using	use	VERB
bracis-19021	134	3	b	b	PROPN
bracis-19021	134	4	initial	initial	ADJ
bracis-19021	134	5	states	state	NOUN
bracis-19021	134	6	sampled	sample	VERB
bracis-19021	134	7	on	on	ADP
bracis-19021	134	8	-	-	PUNCT
bracis-19021	134	9	policy	policy	NOUN
bracis-19021	134	10	(	(	PUNCT
bracis-19021	134	11	\({\mathbf	\({\mathbf	VERB
bracis-19021	134	12	{	{	PUNCT
bracis-19021	134	13	s}}_t	s}}_t	NOUN
bracis-19021	134	14	\sim	\sim	NOUN
bracis-19021	134	15	d^{{\mu	d^{{\mu	NOUN
bracis-19021	134	16	}	}	PUNCT
bracis-19021	134	17	_	_	PUNCT
bracis-19021	134	18	{	{	PUNCT
bracis-19021	134	19	{	{	PUNCT
bracis-19021	134	20	\theta	\theta	PROPN
bracis-19021	134	21	}	}	PUNCT
bracis-19021	134	22	}	}	PUNCT
bracis-19021	134	23	}	}	PUNCT
bracis-19021	134	24	\	\	NOUN
bracis-19021	134	25	)	)	PUNCT
bracis-19021	134	26	)	)	PUNCT
bracis-19021	134	27	and	and	CCONJ
bracis-19021	134	28	k	k	ADJ
bracis-19021	134	29	-	-	PUNCT
bracis-19021	134	30	step	step	NOUN
bracis-19021	134	31	model	model	NOUN
bracis-19021	134	32	-	-	PUNCT
bracis-19021	134	33	based	base	VERB
bracis-19021	134	34	rollouts	rollout	NOUN
bracis-19021	134	35	from	from	ADP
bracis-19021	134	36	each	each	DET
bracis-19021	134	37	state	state	NOUN
bracis-19021	134	38	.	.	PUNCT
bracis-19021	135	1	then	then	ADV
bracis-19021	135	2	,	,	PUNCT
bracis-19021	135	3	we	we	PRON
bracis-19021	135	4	compute	compute	VERB
bracis-19021	135	5	the	the	DET
bracis-19021	135	6	accuracy	accuracy	NOUN
bracis-19021	135	7	as	as	ADP
bracis-19021	135	8	the	the	DET
bracis-19021	135	9	average	average	ADJ
bracis-19021	135	10	cosine	cosine	NOUN
bracis-19021	135	11	similarity	similarity	NOUN
bracis-19021	135	12	of	of	ADP
bracis-19021	135	13	each	each	PRON
bracis-19021	135	14	of	of	ADP
bracis-19021	135	15	the	the	DET
bracis-19021	135	16	10	10	NUM
bracis-19021	135	17	estimates	estimate	NOUN
bracis-19021	135	18	with	with	ADP
bracis-19021	135	19	the	the	DET
bracis-19021	135	20	true	true	ADJ
bracis-19021	135	21	policy	policy	NOUN
bracis-19021	135	22	gradient	gradient	NOUN
bracis-19021	135	23	,	,	PUNCT
bracis-19021	135	24	obtained	obtain	VERB
bracis-19021	135	25	as	as	ADP
bracis-19021	135	26	follows	follow	VERB
bracis-19021	135	27	.	.	PUNCT
bracis-19021	136	1	we	we	PRON
bracis-19021	136	2	first	first	ADV
bracis-19021	136	3	compute	compute	VERB
bracis-19021	136	4	the	the	DET
bracis-19021	136	5	true	true	ADJ
bracis-19021	136	6	expected	expect	VERB
bracis-19021	136	7	return	return	NOUN
bracis-19021	136	8	of	of	ADP
bracis-19021	136	9	a	a	DET
bracis-19021	136	10	policy	policy	NOUN
bracis-19021	136	11	\({\mu	\({\mu	ADP
bracis-19021	136	12	}	}	PUNCT
bracis-19021	136	13	_	_	PUNCT
bracis-19021	136	14	{	{	PUNCT
bracis-19021	136	15	{	{	PUNCT
bracis-19021	136	16	\theta	\theta	PROPN
bracis-19021	136	17	}	}	PUNCT
bracis-19021	136	18	}	}	PUNCT
bracis-19021	136	19	\	\	NOUN
bracis-19021	136	20	)	)	PUNCT
bracis-19021	136	21	via	via	ADP
bracis-19021	136	22	dynamic	dynamic	ADJ
bracis-19021	136	23	programming	programming	NOUN
bracis-19021	136	24	,	,	PUNCT
bracis-19021	136	25	following	follow	VERB
bracis-19021	136	26	eqs	eqs	PROPN
bracis-19021	136	27	.	.	PUNCT
bracis-19021	137	1	(	(	PUNCT
bracis-19021	137	2	4	4	NUM
bracis-19021	137	3	)	)	PUNCT
bracis-19021	137	4	and	and	CCONJ
bracis-19021	137	5	(	(	PUNCT
bracis-19021	137	6	5	5	NUM
bracis-19021	137	7	)	)	PUNCT
bracis-19021	137	8	.	.	PUNCT
bracis-19021	138	1	our	our	PRON
bracis-19021	138	2	implementation	implementation	NOUN
bracis-19021	138	3	in	in	ADP
bracis-19021	138	4	pytorch	pytorch	NOUN
bracis-19021	138	5	[	[	X
bracis-19021	138	6	18	18	NUM
bracis-19021	138	7	]	]	PUNCT
bracis-19021	138	8	then	then	ADV
bracis-19021	138	9	allows	allow	VERB
bracis-19021	138	10	us	we	PRON
bracis-19021	138	11	to	to	PART
bracis-19021	138	12	use	use	VERB
bracis-19021	138	13	automatic	automatic	ADJ
bracis-19021	138	14	differentiation	differentiation	NOUN
bracis-19021	138	15	to	to	PART
bracis-19021	138	16	compute	compute	VERB
bracis-19021	138	17	the	the	DET
bracis-19021	138	18	gradient	gradient	NOUN
bracis-19021	138	19	of	of	ADP
bracis-19021	138	20	the	the	DET
bracis-19021	138	21	expected	expect	VERB
bracis-19021	138	22	return	return	NOUN
bracis-19021	138	23	w.r.t	w.r.t	VERB
bracis-19021	138	24	.	.	PUNCT
bracis-19021	139	1	policy	policy	NOUN
bracis-19021	139	2	parameters	parameter	NOUN
bracis-19021	139	3	.	.	PUNCT
bracis-19021	140	1	the	the	DET
bracis-19021	140	2	second	second	ADJ
bracis-19021	140	3	metric	metric	NOUN
bracis-19021	140	4	is	be	AUX
bracis-19021	140	5	a	a	DET
bracis-19021	140	6	measure	measure	NOUN
bracis-19021	140	7	of	of	ADP
bracis-19021	140	8	gradient	gradient	ADJ
bracis-19021	140	9	precision	precision	NOUN
bracis-19021	140	10	and	and	CCONJ
bracis-19021	140	11	we	we	PRON
bracis-19021	140	12	denote	denote	VERB
bracis-19021	140	13	it	it	PRON
bracis-19021	140	14	as	as	ADP
bracis-19021	140	15	such	such	ADJ
bracis-19021	140	16	in	in	ADP
bracis-19021	140	17	the	the	DET
bracis-19021	140	18	following	follow	VERB
bracis-19021	140	19	plots	plot	NOUN
bracis-19021	140	20	.	.	PUNCT
bracis-19021	141	1	again	again	ADV
bracis-19021	141	2	,	,	PUNCT
bracis-19021	141	3	we	we	PRON
bracis-19021	141	4	compute	compute	VERB
bracis-19021	141	5	10	10	NUM
bracis-19021	141	6	estimates	estimate	NOUN
bracis-19021	141	7	of	of	ADP
bracis-19021	141	8	the	the	DET
bracis-19021	141	9	gradient	gradient	NOUN
bracis-19021	141	10	in	in	ADP
bracis-19021	141	11	the	the	DET
bracis-19021	141	12	same	same	ADJ
bracis-19021	141	13	manner	manner	NOUN
bracis-19021	141	14	used	use	VERB
bracis-19021	141	15	in	in	ADP
bracis-19021	141	16	computing	compute	VERB
bracis-19021	141	17	the	the	DET
bracis-19021	141	18	accuracy	accuracy	NOUN
bracis-19021	141	19	.	.	PUNCT
bracis-19021	142	1	then	then	ADV
bracis-19021	142	2	,	,	PUNCT
bracis-19021	142	3	we	we	PRON
bracis-19021	142	4	compute	compute	VERB
bracis-19021	142	5	the	the	DET
bracis-19021	142	6	precision	precision	NOUN
bracis-19021	142	7	as	as	ADP
bracis-19021	142	8	the	the	DET
bracis-19021	142	9	average	average	ADJ
bracis-19021	142	10	pairwise	pairwise	NOUN
bracis-19021	142	11	cosine	cosine	NOUN
bracis-19021	142	12	similarity	similarity	NOUN
bracis-19021	142	13	of	of	ADP
bracis-19021	142	14	the	the	DET
bracis-19021	142	15	10	10	NUM
bracis-19021	142	16	estimates	estimate	NOUN
bracis-19021	142	17	(	(	PUNCT
bracis-19021	142	18	the	the	PRON
bracis-19021	142	19	higher	high	ADJ
bracis-19021	142	20	this	this	DET
bracis-19021	142	21	quantity	quantity	NOUN
bracis-19021	142	22	,	,	PUNCT
bracis-19021	142	23	the	the	PRON
bracis-19021	142	24	lower	low	ADJ
bracis-19021	142	25	the	the	DET
bracis-19021	142	26	variance	variance	NOUN
bracis-19021	142	27	)	)	PUNCT
bracis-19021	142	28	.	.	PUNCT
bracis-19021	143	1	fig	fig	NOUN
bracis-19021	143	2	.	.	PUNCT
bracis-19021	144	1	2	2	NUM
bracis-19021	144	2	.	.	X
bracis-19021	144	3	gradient	gradient	ADJ
bracis-19021	144	4	accuracy	accuracy	NOUN
bracis-19021	144	5	for	for	ADP
bracis-19021	144	6	each	each	DET
bracis-19021	144	7	estimator	estimator	NOUN
bracis-19021	144	8	near	near	ADP
bracis-19021	144	9	convergence	convergence	NOUN
bracis-19021	144	10	for	for	ADP
bracis-19021	144	11	different	different	ADJ
bracis-19021	144	12	virtual	virtual	ADJ
bracis-19021	144	13	rollout	rollout	NOUN
bracis-19021	144	14	lengths	length	NOUN
bracis-19021	144	15	(	(	PUNCT
bracis-19021	144	16	k	k	NOUN
bracis-19021	144	17	)	)	PUNCT
bracis-19021	144	18	.	.	PUNCT
bracis-19021	145	1	we	we	PRON
bracis-19021	145	2	used	use	VERB
bracis-19021	145	3	50000	50000	NUM
bracis-19021	145	4	states	state	NOUN
bracis-19021	145	5	sampled	sample	VERB
bracis-19021	145	6	from	from	ADP
bracis-19021	145	7	the	the	DET
bracis-19021	145	8	policy	policy	NOUN
bracis-19021	145	9	to	to	PART
bracis-19021	145	10	approximate	approximate	VERB
bracis-19021	145	11	the	the	DET
bracis-19021	145	12	expected	expect	VERB
bracis-19021	145	13	value	value	NOUN
bracis-19021	145	14	.	.	PUNCT
bracis-19021	146	1	full	full	ADJ
bracis-19021	146	2	size	size	NOUN
bracis-19021	146	3	image	image	NOUN
bracis-19021	146	4	we	we	PRON
bracis-19021	146	5	first	first	ADV
bracis-19021	146	6	analyze	analyze	VERB
bracis-19021	146	7	the	the	DET
bracis-19021	146	8	accuracy	accuracy	NOUN
bracis-19021	146	9	of	of	ADP
bracis-19021	146	10	each	each	DET
bracis-19021	146	11	estimator	estimator	NOUN
bracis-19021	146	12	when	when	SCONJ
bracis-19021	146	13	given	give	VERB
bracis-19021	146	14	enough	enough	ADJ
bracis-19021	146	15	states	state	NOUN
bracis-19021	146	16	from	from	ADP
bracis-19021	146	17	the	the	DET
bracis-19021	146	18	policy	policy	NOUN
bracis-19021	146	19	’s	’s	PART
bracis-19021	146	20	distribution	distribution	NOUN
bracis-19021	146	21	\(d^{{\mu	\(d^{{\mu	NOUN
bracis-19021	146	22	}	}	PUNCT
bracis-19021	146	23	_	_	PUNCT
bracis-19021	146	24	{	{	PUNCT
bracis-19021	146	25	{	{	PUNCT
bracis-19021	146	26	\theta	\theta	PROPN
bracis-19021	146	27	}	}	PUNCT
bracis-19021	146	28	}	}	PUNCT
bracis-19021	146	29	}	}	PUNCT
bracis-19021	146	30	\	\	NOUN
bracis-19021	146	31	)	)	PUNCT
bracis-19021	146	32	to	to	PART
bracis-19021	146	33	approximate	approximate	VERB
bracis-19021	146	34	their	their	PRON
bracis-19021	146	35	true	true	ADJ
bracis-19021	146	36	expected	expected	ADJ
bracis-19021	146	37	values	value	NOUN
bracis-19021	146	38	.	.	PUNCT
bracis-19021	147	1	figure	figure	VERB
bracis-19021	147	2	 	 	SPACE
bracis-19021	147	3	2	2	NUM
bracis-19021	147	4	shows	show	VERB
bracis-19021	147	5	the	the	DET
bracis-19021	147	6	accuracy	accuracy	NOUN
bracis-19021	147	7	obtained	obtain	VERB
bracis-19021	147	8	by	by	ADP
bracis-19021	147	9	dpg	dpg	PROPN
bracis-19021	147	10	and	and	CCONJ
bracis-19021	147	11	maac	maac	NOUN
bracis-19021	147	12	for	for	ADP
bracis-19021	147	13	different	different	ADJ
bracis-19021	147	14	values	value	NOUN
bracis-19021	147	15	of	of	ADP
bracis-19021	147	16	k	k	PROPN
bracis-19021	147	17	using	use	VERB
bracis-19021	147	18	50000	50000	NUM
bracis-19021	147	19	states	state	NOUN
bracis-19021	147	20	from	from	ADP
bracis-19021	147	21	the	the	DET
bracis-19021	147	22	policy	policy	NOUN
bracis-19021	147	23	’s	’s	PART
bracis-19021	147	24	distribution	distribution	NOUN
bracis-19021	147	25	.	.	PUNCT
bracis-19021	148	1	the	the	DET
bracis-19021	148	2	lqgs	lqg	NOUN
bracis-19021	148	3	considered	consider	VERB
bracis-19021	148	4	have	have	VERB
bracis-19021	148	5	state	state	NOUN
bracis-19021	148	6	and	and	CCONJ
bracis-19021	148	7	action	action	NOUN
bracis-19021	148	8	spaces	space	NOUN
bracis-19021	148	9	of	of	ADP
bracis-19021	148	10	dimension	dimension	NOUN
bracis-19021	148	11	2	2	NUM
bracis-19021	148	12	and	and	CCONJ
bracis-19021	148	13	horizon	horizon	NOUN
bracis-19021	148	14	of	of	ADP
bracis-19021	148	15	length	length	NOUN
bracis-19021	148	16	20	20	NUM
bracis-19021	148	17	.	.	PUNCT
bracis-19021	149	1	for	for	ADP
bracis-19021	149	2	each	each	DET
bracis-19021	149	3	value	value	NOUN
bracis-19021	149	4	of	of	ADP
bracis-19021	149	5	k	k	PROPN
bracis-19021	149	6	,	,	PUNCT
bracis-19021	149	7	we	we	PRON
bracis-19021	149	8	initialize	initialize	VERB
bracis-19021	149	9	10	10	NUM
bracis-19021	149	10	different	different	ADJ
bracis-19021	149	11	environment	environment	NOUN
bracis-19021	149	12	-	-	PUNCT
bracis-19021	149	13	policy	policy	NOUN
bracis-19021	149	14	pairs	pair	NOUN
bracis-19021	149	15	and	and	CCONJ
bracis-19021	149	16	compute	compute	VERB
bracis-19021	149	17	the	the	DET
bracis-19021	149	18	accuracy	accuracy	NOUN
bracis-19021	149	19	for	for	ADP
bracis-19021	149	20	each	each	PRON
bracis-19021	149	21	,	,	PUNCT
bracis-19021	149	22	denoted	denote	VERB
bracis-19021	149	23	as	as	ADP
bracis-19021	149	24	different	different	ADJ
bracis-19021	149	25	markers	marker	NOUN
bracis-19021	149	26	in	in	ADP
bracis-19021	149	27	each	each	PRON
bracis-19021	149	28	vertical	vertical	ADJ
bracis-19021	149	29	line.footnote	line.footnote	NOUN
bracis-19021	149	30	4	4	NUM
bracis-19021	149	31	note	note	VERB
bracis-19021	149	32	how	how	SCONJ
bracis-19021	149	33	all	all	PRON
bracis-19021	149	34	but	but	SCONJ
bracis-19021	149	35	one	one	NUM
bracis-19021	149	36	of	of	ADP
bracis-19021	149	37	the	the	DET
bracis-19021	149	38	instances	instance	NOUN
bracis-19021	149	39	using	use	VERB
bracis-19021	149	40	dpg(k	dpg(k	NOUN
bracis-19021	149	41	)	)	PUNCT
bracis-19021	149	42	converged	converge	VERB
bracis-19021	149	43	to	to	ADP
bracis-19021	149	44	the	the	DET
bracis-19021	149	45	true	true	ADJ
bracis-19021	149	46	value	value	NOUN
bracis-19021	149	47	gradient	gradient	NOUN
bracis-19021	149	48	,	,	PUNCT
bracis-19021	149	49	indicating	indicate	VERB
bracis-19021	149	50	that	that	SCONJ
bracis-19021	149	51	it	it	PRON
bracis-19021	149	52	is	be	AUX
bracis-19021	149	53	indeed	indeed	ADV
bracis-19021	149	54	an	an	DET
bracis-19021	149	55	unbiased	unbiased	ADJ
bracis-19021	149	56	estimator	estimator	NOUN
bracis-19021	149	57	.	.	PUNCT
bracis-19021	150	1	on	on	ADP
bracis-19021	150	2	the	the	DET
bracis-19021	150	3	other	other	ADJ
bracis-19021	150	4	hand	hand	NOUN
bracis-19021	150	5	,	,	PUNCT
bracis-19021	150	6	maac(k	maac(k	NOUN
bracis-19021	150	7	)	)	PUNCT
bracis-19021	150	8	incurs	incur	VERB
bracis-19021	150	9	a	a	DET
bracis-19021	150	10	larger	large	ADJ
bracis-19021	150	11	bias	bias	NOUN
bracis-19021	150	12	with	with	ADP
bracis-19021	150	13	increasing	increase	VERB
bracis-19021	150	14	values	value	NOUN
bracis-19021	150	15	of	of	ADP
bracis-19021	150	16	k	k	NOUN
bracis-19021	150	17	,	,	PUNCT
bracis-19021	150	18	indicating	indicate	VERB
bracis-19021	150	19	that	that	SCONJ
bracis-19021	150	20	the	the	DET
bracis-19021	150	21	added	add	VERB
bracis-19021	150	22	action	action	NOUN
bracis-19021	150	23	action	action	NOUN
bracis-19021	150	24	-	-	PUNCT
bracis-19021	150	25	gradient	gradient	NOUN
bracis-19021	150	26	terms	term	NOUN
bracis-19021	150	27	(	(	PUNCT
bracis-19021	150	28	see	see	VERB
bracis-19021	150	29	fig	fig	NOUN
bracis-19021	150	30	.	.	PUNCT
bracis-19021	150	31	 	 	SPACE
bracis-19021	150	32	1	1	X
bracis-19021	150	33	)	)	PUNCT
bracis-19021	150	34	influence	influence	NOUN
bracis-19021	150	35	the	the	DET
bracis-19021	150	36	final	final	ADJ
bracis-19021	150	37	direction	direction	NOUN
bracis-19021	150	38	.	.	PUNCT
bracis-19021	151	1	although	although	SCONJ
bracis-19021	151	2	the	the	DET
bracis-19021	151	3	results	result	NOUN
bracis-19021	151	4	above	above	ADP
bracis-19021	151	5	indicate	indicate	VERB
bracis-19021	151	6	that	that	SCONJ
bracis-19021	151	7	maac(k	maac(k	NOUN
bracis-19021	151	8	)	)	PUNCT
bracis-19021	151	9	is	be	AUX
bracis-19021	151	10	biased	bias	VERB
bracis-19021	151	11	at	at	ADP
bracis-19021	151	12	convergence	convergence	NOUN
bracis-19021	151	13	,	,	PUNCT
bracis-19021	151	14	most	most	ADJ
bracis-19021	151	15	svg	svg	NOUN
bracis-19021	151	16	algorithms	algorithm	NOUN
bracis-19021	151	17	operate	operate	VERB
bracis-19021	151	18	on	on	ADP
bracis-19021	151	19	a	a	DET
bracis-19021	151	20	much	much	ADV
bracis-19021	151	21	smaller	small	ADJ
bracis-19021	151	22	sample	sample	NOUN
bracis-19021	151	23	regime	regime	NOUN
bracis-19021	151	24	.	.	PUNCT
bracis-19021	152	1	figure	figure	NOUN
bracis-19021	152	2	 	 	SPACE
bracis-19021	152	3	3	3	NUM
bracis-19021	152	4	shows	show	VERB
bracis-19021	152	5	the	the	DET
bracis-19021	152	6	accuracy	accuracy	NOUN
bracis-19021	152	7	across	across	ADP
bracis-19021	152	8	10	10	NUM
bracis-19021	152	9	different	different	ADJ
bracis-19021	152	10	environment	environment	NOUN
bracis-19021	152	11	-	-	PUNCT
bracis-19021	152	12	policy	policy	NOUN
bracis-19021	152	13	pairs	pair	NOUN
bracis-19021	152	14	;	;	PUNCT
bracis-19021	152	15	this	this	DET
bracis-19021	152	16	time	time	NOUN
bracis-19021	152	17	,	,	PUNCT
bracis-19021	152	18	however	however	ADV
bracis-19021	152	19	,	,	PUNCT
bracis-19021	152	20	using	use	VERB
bracis-19021	152	21	smaller	small	ADJ
bracis-19021	152	22	sample	sample	NOUN
bracis-19021	152	23	sizes	size	NOUN
bracis-19021	152	24	from	from	ADP
bracis-19021	152	25	the	the	DET
bracis-19021	152	26	policy	policy	NOUN
bracis-19021	152	27	distribution	distribution	NOUN
bracis-19021	152	28	.	.	PUNCT
bracis-19021	153	1	lines	line	NOUN
bracis-19021	153	2	denote	denote	VERB
bracis-19021	153	3	the	the	DET
bracis-19021	153	4	average	average	ADJ
bracis-19021	153	5	results	result	NOUN
bracis-19021	153	6	and	and	CCONJ
bracis-19021	153	7	shaded	shaded	ADJ
bracis-19021	153	8	areas	area	NOUN
bracis-19021	153	9	,	,	PUNCT
bracis-19021	153	10	the	the	DET
bracis-19021	153	11	95	95	NUM
bracis-19021	153	12	%	%	NOUN
bracis-19021	153	13	confidence	confidence	NOUN
bracis-19021	153	14	interval.footnote	interval.footnote	PROPN
bracis-19021	153	15	5	5	NUM
bracis-19021	153	16	for	for	ADP
bracis-19021	153	17	\(k=0\	\(k=0\	NOUN
bracis-19021	153	18	)	)	PUNCT
bracis-19021	153	19	,	,	PUNCT
bracis-19021	153	20	the	the	DET
bracis-19021	153	21	estimators	estimator	NOUN
bracis-19021	153	22	are	be	AUX
bracis-19021	153	23	equivalent	equivalent	ADJ
bracis-19021	153	24	,	,	PUNCT
bracis-19021	153	25	which	which	PRON
bracis-19021	153	26	is	be	AUX
bracis-19021	153	27	verified	verify	VERB
bracis-19021	153	28	in	in	ADP
bracis-19021	153	29	practice	practice	NOUN
bracis-19021	153	30	.	.	PUNCT
bracis-19021	154	1	in	in	ADP
bracis-19021	154	2	this	this	DET
bracis-19021	154	3	more	more	ADV
bracis-19021	154	4	practical	practical	ADJ
bracis-19021	154	5	sample	sample	NOUN
bracis-19021	154	6	regime	regime	NOUN
bracis-19021	154	7	,	,	PUNCT
bracis-19021	154	8	we	we	PRON
bracis-19021	154	9	see	see	VERB
bracis-19021	154	10	that	that	SCONJ
bracis-19021	154	11	maac(k	maac(k	NOUN
bracis-19021	154	12	)	)	PUNCT
bracis-19021	154	13	produces	produce	VERB
bracis-19021	154	14	more	more	ADV
bracis-19021	154	15	accurate	accurate	ADJ
bracis-19021	154	16	results	result	NOUN
bracis-19021	154	17	,	,	PUNCT
bracis-19021	154	18	specially	specially	ADV
bracis-19021	154	19	for	for	ADP
bracis-19021	154	20	larger	large	ADJ
bracis-19021	154	21	values	value	NOUN
bracis-19021	154	22	of	of	ADP
bracis-19021	154	23	k.	k.	PROPN
bracis-19021	154	24	similar	similar	ADJ
bracis-19021	154	25	to	to	ADP
bracis-19021	154	26	fig	fig	NOUN
bracis-19021	154	27	.	.	PUNCT
bracis-19021	154	28	 	 	SPACE
bracis-19021	155	1	3	3	NUM
bracis-19021	155	2	,	,	PUNCT
bracis-19021	155	3	fig	fig	NOUN
bracis-19021	155	4	.	.	PUNCT
bracis-19021	155	5	 	 	SPACE
bracis-19021	156	1	4	4	NUM
bracis-19021	156	2	shows	show	VERB
bracis-19021	156	3	the	the	DET
bracis-19021	156	4	gradient	gradient	ADJ
bracis-19021	156	5	precision	precision	NOUN
bracis-19021	156	6	in	in	ADP
bracis-19021	156	7	the	the	DET
bracis-19021	156	8	same	same	ADJ
bracis-19021	156	9	setting	setting	NOUN
bracis-19021	156	10	.	.	PUNCT
bracis-19021	157	1	we	we	PRON
bracis-19021	157	2	see	see	VERB
bracis-19021	157	3	that	that	SCONJ
bracis-19021	157	4	the	the	DET
bracis-19021	157	5	variance	variance	NOUN
bracis-19021	157	6	of	of	ADP
bracis-19021	157	7	maac(k	maac(k	NOUN
bracis-19021	157	8	)	)	PUNCT
bracis-19021	157	9	is	be	AUX
bracis-19021	157	10	lower	low	ADJ
bracis-19021	157	11	than	than	ADP
bracis-19021	157	12	that	that	PRON
bracis-19021	157	13	of	of	ADP
bracis-19021	157	14	dpg(k	dpg(k	NOUN
bracis-19021	157	15	)	)	PUNCT
bracis-19021	157	16	across	across	ADP
bracis-19021	157	17	all	all	DET
bracis-19021	157	18	tested	test	VERB
bracis-19021	157	19	values	value	NOUN
bracis-19021	157	20	of	of	ADP
bracis-19021	157	21	\(k>0\	\(k>0\	NOUN
bracis-19021	157	22	)	)	PUNCT
bracis-19021	157	23	.	.	PUNCT
bracis-19021	158	1	overall	overall	ADJ
bracis-19021	158	2	,	,	PUNCT
bracis-19021	158	3	figs	fig	NOUN
bracis-19021	158	4	.	.	PUNCT
bracis-19021	158	5	 	 	SPACE
bracis-19021	158	6	2	2	NUM
bracis-19021	158	7	,	,	PUNCT
bracis-19021	158	8	3	3	NUM
bracis-19021	158	9	and	and	CCONJ
bracis-19021	158	10	4	4	NUM
bracis-19021	158	11	illustrate	illustrate	VERB
bracis-19021	158	12	a	a	DET
bracis-19021	158	13	classic	classic	ADJ
bracis-19021	158	14	instance	instance	NOUN
bracis-19021	158	15	of	of	ADP
bracis-19021	158	16	the	the	DET
bracis-19021	158	17	bias	bias	NOUN
bracis-19021	158	18	-	-	PUNCT
bracis-19021	158	19	variance	variance	NOUN
bracis-19021	158	20	tradeoff	tradeoff	NOUN
bracis-19021	158	21	in	in	ADP
bracis-19021	158	22	machine	machine	NOUN
bracis-19021	158	23	learning	learning	NOUN
bracis-19021	158	24	:	:	PUNCT
bracis-19021	158	25	maac(k	maac(k	ADJ
bracis-19021	158	26	)	)	PUNCT
bracis-19021	158	27	introduces	introduce	NOUN
bracis-19021	158	28	bias	bias	NOUN
bracis-19021	158	29	,	,	PUNCT
bracis-19021	158	30	although	although	SCONJ
bracis-19021	158	31	a	a	DET
bracis-19021	158	32	small	small	ADJ
bracis-19021	158	33	one	one	NOUN
bracis-19021	158	34	,	,	PUNCT
bracis-19021	158	35	in	in	ADP
bracis-19021	158	36	return	return	NOUN
bracis-19021	158	37	for	for	ADP
bracis-19021	158	38	a	a	DET
bracis-19021	158	39	much	much	ADV
bracis-19021	158	40	more	more	ADV
bracis-19021	158	41	stable	stable	ADJ
bracis-19021	158	42	(	(	PUNCT
bracis-19021	158	43	less	less	ADV
bracis-19021	158	44	variable	variable	ADJ
bracis-19021	158	45	)	)	PUNCT
bracis-19021	158	46	estimate	estimate	NOUN
bracis-19021	158	47	of	of	ADP
bracis-19021	158	48	the	the	DET
bracis-19021	158	49	gradient	gradient	NOUN
bracis-19021	158	50	,	,	PUNCT
bracis-19021	158	51	whereas	whereas	SCONJ
bracis-19021	158	52	the	the	DET
bracis-19021	158	53	unbiased	unbiased	ADJ
bracis-19021	158	54	dpg(k	dpg(k	NOUN
bracis-19021	158	55	)	)	PUNCT
bracis-19021	158	56	demands	demand	VERB
bracis-19021	158	57	much	much	ADV
bracis-19021	158	58	more	more	ADJ
bracis-19021	158	59	samples	sample	NOUN
bracis-19021	158	60	to	to	PART
bracis-19021	158	61	justify	justify	VERB
bracis-19021	158	62	its	its	PRON
bracis-19021	158	63	use	use	NOUN
bracis-19021	158	64	.	.	PUNCT
bracis-19021	159	1	fig	fig	NOUN
bracis-19021	159	2	.	.	PUNCT
bracis-19021	160	1	3	3	X
bracis-19021	160	2	.	.	X
bracis-19021	160	3	gradient	gradient	ADJ
bracis-19021	160	4	accuracy	accuracy	NOUN
bracis-19021	160	5	for	for	ADP
bracis-19021	160	6	each	each	DET
bracis-19021	160	7	estimator	estimator	NOUN
bracis-19021	160	8	for	for	ADP
bracis-19021	160	9	different	different	ADJ
bracis-19021	160	10	minibatch	minibatch	NOUN
bracis-19021	160	11	sizes	size	NOUN
bracis-19021	160	12	(	(	PUNCT
bracis-19021	160	13	\(b	\(b	NOUN
bracis-19021	160	14	=	=	SYM
bracis-19021	160	15	\	\	NOUN
bracis-19021	160	16	)	)	PUNCT
bracis-19021	160	17	#	#	NOUN
bracis-19021	160	18	samples	sample	NOUN
bracis-19021	160	19	)	)	PUNCT
bracis-19021	160	20	and	and	CCONJ
bracis-19021	160	21	virtual	virtual	ADJ
bracis-19021	160	22	rollout	rollout	NOUN
bracis-19021	160	23	lengths	length	NOUN
bracis-19021	160	24	(	(	PUNCT
bracis-19021	160	25	k	k	NOUN
bracis-19021	160	26	)	)	PUNCT
bracis-19021	160	27	.	.	PUNCT
bracis-19021	161	1	full	full	ADJ
bracis-19021	161	2	size	size	NOUN
bracis-19021	161	3	image	image	NOUN
bracis-19021	161	4	note	note	VERB
bracis-19021	161	5	that	that	SCONJ
bracis-19021	161	6	the	the	DET
bracis-19021	161	7	accuracy	accuracy	NOUN
bracis-19021	161	8	and	and	CCONJ
bracis-19021	161	9	precision	precision	NOUN
bracis-19021	161	10	metrics	metric	NOUN
bracis-19021	161	11	only	only	ADV
bracis-19021	161	12	account	account	VERB
bracis-19021	161	13	for	for	ADP
bracis-19021	161	14	differences	difference	NOUN
bracis-19021	161	15	in	in	ADP
bracis-19021	161	16	gradient	gradient	ADJ
bracis-19021	161	17	direction	direction	NOUN
bracis-19021	161	18	and	and	CCONJ
bracis-19021	161	19	orientation	orientation	NOUN
bracis-19021	161	20	.	.	PUNCT
bracis-19021	162	1	the	the	DET
bracis-19021	162	2	magnitude	magnitude	NOUN
bracis-19021	162	3	may	may	AUX
bracis-19021	162	4	also	also	ADV
bracis-19021	162	5	be	be	AUX
bracis-19021	162	6	important	important	ADJ
bracis-19021	162	7	,	,	PUNCT
bracis-19021	162	8	as	as	SCONJ
bracis-19021	162	9	it	it	PRON
bracis-19021	162	10	influences	influence	VERB
bracis-19021	162	11	the	the	DET
bracis-19021	162	12	learning	learning	NOUN
bracis-19021	162	13	rate	rate	NOUN
bracis-19021	162	14	when	when	SCONJ
bracis-19021	162	15	used	use	VERB
bracis-19021	162	16	to	to	PART
bracis-19021	162	17	update	update	VERB
bracis-19021	162	18	policy	policy	NOUN
bracis-19021	162	19	parameters	parameter	NOUN
bracis-19021	162	20	.	.	PUNCT
bracis-19021	163	1	figure	figure	NOUN
bracis-19021	163	2	 	 	SPACE
bracis-19021	163	3	5	5	NUM
bracis-19021	163	4	shows	show	VERB
bracis-19021	163	5	that	that	SCONJ
bracis-19021	163	6	maac(k	maac(k	NOUN
bracis-19021	163	7	)	)	PUNCT
bracis-19021	163	8	produces	produce	VERB
bracis-19021	163	9	gradients	gradient	NOUN
bracis-19021	163	10	with	with	ADP
bracis-19021	163	11	higher	high	ADJ
bracis-19021	163	12	norms	norm	NOUN
bracis-19021	163	13	compared	compare	VERB
bracis-19021	163	14	to	to	ADP
bracis-19021	163	15	dpg(k	dpg(k	NOUN
bracis-19021	163	16	)	)	PUNCT
bracis-19021	163	17	.	.	PUNCT
bracis-19021	164	1	one	one	PRON
bracis-19021	164	2	should	should	AUX
bracis-19021	164	3	keep	keep	VERB
bracis-19021	164	4	this	this	PRON
bracis-19021	164	5	in	in	ADP
bracis-19021	164	6	mind	mind	NOUN
bracis-19021	164	7	when	when	SCONJ
bracis-19021	164	8	choosing	choose	VERB
bracis-19021	164	9	the	the	DET
bracis-19021	164	10	learning	learning	NOUN
bracis-19021	164	11	rate	rate	NOUN
bracis-19021	164	12	for	for	ADP
bracis-19021	164	13	sgd	sgd	NOUN
bracis-19021	164	14	,	,	PUNCT
bracis-19021	164	15	as	as	SCONJ
bracis-19021	164	16	the	the	DET
bracis-19021	164	17	following	follow	VERB
bracis-19021	164	18	experiments	experiment	NOUN
bracis-19021	164	19	show	show	VERB
bracis-19021	164	20	that	that	SCONJ
bracis-19021	164	21	the	the	DET
bracis-19021	164	22	gradient	gradient	NOUN
bracis-19021	164	23	norm	norm	NOUN
bracis-19021	164	24	have	have	VERB
bracis-19021	164	25	a	a	DET
bracis-19021	164	26	significant	significant	ADJ
bracis-19021	164	27	impact	impact	NOUN
bracis-19021	164	28	on	on	ADP
bracis-19021	164	29	policy	policy	NOUN
bracis-19021	164	30	optimization	optimization	NOUN
bracis-19021	164	31	.	.	PUNCT
bracis-19021	165	1	5.2	5.2	NUM
bracis-19021	165	2	impact	impact	NOUN
bracis-19021	165	3	of	of	ADP
bracis-19021	165	4	gradient	gradient	ADJ
bracis-19021	165	5	quality	quality	NOUN
bracis-19021	165	6	on	on	ADP
bracis-19021	165	7	policy	policy	NOUN
bracis-19021	165	8	optimization	optimization	NOUN
bracis-19021	165	9	previous	previous	ADJ
bracis-19021	165	10	work	work	NOUN
bracis-19021	165	11	on	on	ADP
bracis-19021	165	12	model	model	ADJ
bracis-19021	165	13	-	-	PUNCT
bracis-19021	165	14	free	free	ADJ
bracis-19021	165	15	policy	policy	NOUN
bracis-19021	165	16	gradients	gradient	NOUN
bracis-19021	165	17	has	have	AUX
bracis-19021	165	18	shown	show	VERB
bracis-19021	165	19	that	that	SCONJ
bracis-19021	165	20	policy	policy	NOUN
bracis-19021	165	21	optimization	optimization	NOUN
bracis-19021	165	22	algorithms	algorithm	NOUN
bracis-19021	165	23	can	can	AUX
bracis-19021	165	24	improve	improve	VERB
bracis-19021	165	25	a	a	DET
bracis-19021	165	26	policy	policy	NOUN
bracis-19021	165	27	despite	despite	SCONJ
bracis-19021	165	28	using	use	VERB
bracis-19021	165	29	poor	poor	ADJ
bracis-19021	165	30	gradient	gradient	ADJ
bracis-19021	165	31	estimation	estimation	NOUN
bracis-19021	165	32	[	[	X
bracis-19021	165	33	11	11	NUM
bracis-19021	165	34	]	]	PUNCT
bracis-19021	165	35	.	.	PUNCT
bracis-19021	166	1	it	it	PRON
bracis-19021	166	2	is	be	AUX
bracis-19021	166	3	not	not	PART
bracis-19021	166	4	clear	clear	ADJ
bracis-19021	166	5	,	,	PUNCT
bracis-19021	166	6	however	however	ADV
bracis-19021	166	7	,	,	PUNCT
bracis-19021	166	8	if	if	SCONJ
bracis-19021	166	9	better	well	ADJ
bracis-19021	166	10	policy	policy	NOUN
bracis-19021	166	11	gradient	gradient	NOUN
bracis-19021	166	12	estimation	estimation	NOUN
bracis-19021	166	13	translates	translate	VERB
bracis-19021	166	14	to	to	ADP
bracis-19021	166	15	more	more	ADJ
bracis-19021	166	16	stability	stability	NOUN
bracis-19021	166	17	or	or	CCONJ
bracis-19021	166	18	faster	fast	ADJ
bracis-19021	166	19	convergence	convergence	NOUN
bracis-19021	166	20	in	in	ADP
bracis-19021	166	21	svg	svg	PROPN
bracis-19021	166	22	algorithms	algorithm	NOUN
bracis-19021	166	23	.	.	PUNCT
bracis-19021	167	1	we	we	PRON
bracis-19021	167	2	therefore	therefore	ADV
bracis-19021	167	3	conduct	conduct	VERB
bracis-19021	167	4	our	our	PRON
bracis-19021	167	5	next	next	ADJ
bracis-19021	167	6	experiments	experiment	NOUN
bracis-19021	167	7	comparing	compare	VERB
bracis-19021	167	8	the	the	DET
bracis-19021	167	9	maac	maac	NOUN
bracis-19021	167	10	and	and	CCONJ
bracis-19021	167	11	dpg	dpg	ADJ
bracis-19021	167	12	estimators	estimator	NOUN
bracis-19021	167	13	for	for	ADP
bracis-19021	167	14	policy	policy	NOUN
bracis-19021	167	15	optimization	optimization	NOUN
bracis-19021	167	16	.	.	PUNCT
bracis-19021	168	1	figures	figure	NOUN
bracis-19021	168	2	 	 	SPACE
bracis-19021	168	3	6	6	NUM
bracis-19021	168	4	and	and	CCONJ
bracis-19021	168	5	7	7	NUM
bracis-19021	168	6	show	show	NOUN
bracis-19021	168	7	learning	learn	VERB
bracis-19021	168	8	curves	curve	NOUN
bracis-19021	168	9	as	as	ADP
bracis-19021	168	10	total	total	ADJ
bracis-19021	168	11	cost	cost	NOUN
bracis-19021	168	12	(	(	PUNCT
bracis-19021	168	13	negative	negative	ADJ
bracis-19021	168	14	return	return	NOUN
bracis-19021	168	15	)	)	PUNCT
bracis-19021	168	16	against	against	ADP
bracis-19021	168	17	the	the	DET
bracis-19021	168	18	number	number	NOUN
bracis-19021	168	19	of	of	ADP
bracis-19021	168	20	sgd	sgd	PROPN
bracis-19021	168	21	iterations	iteration	NOUN
bracis-19021	168	22	across	across	ADP
bracis-19021	168	23	several	several	ADJ
bracis-19021	168	24	instances	instance	NOUN
bracis-19021	168	25	of	of	ADP
bracis-19021	168	26	lqgs	lqg	NOUN
bracis-19021	168	27	(	(	PUNCT
bracis-19021	168	28	\(\text	\(\text	NOUN
bracis-19021	168	29	{	{	PUNCT
bracis-19021	168	30	dim}({\mathcal	dim}({\mathcal	PROPN
bracis-19021	168	31	s	s	NOUN
bracis-19021	168	32	}	}	PUNCT
bracis-19021	168	33	)	)	PUNCT
bracis-19021	169	1	=	=	SYM
bracis-19021	169	2	\text	\text	ADJ
bracis-19021	169	3	{	{	PUNCT
bracis-19021	169	4	dim}({\mathcal	dim}({\mathcal	ADJ
bracis-19021	169	5	a	a	NOUN
bracis-19021	169	6	}	}	PUNCT
bracis-19021	169	7	)	)	PUNCT
bracis-19021	169	8	=	=	SYM
bracis-19021	169	9	2\	2\	NUM
bracis-19021	169	10	)	)	PUNCT
bracis-19021	169	11	and	and	CCONJ
bracis-19021	169	12	\(h=	\(h=	NUM
bracis-19021	169	13	20\	20\	NUM
bracis-19021	169	14	)	)	PUNCT
bracis-19021	169	15	)	)	PUNCT
bracis-19021	169	16	.	.	PUNCT
bracis-19021	170	1	we	we	PRON
bracis-19021	170	2	use	use	VERB
bracis-19021	170	3	the	the	DET
bracis-19021	170	4	same	same	ADJ
bracis-19021	170	5	hyperparameters	hyperparameter	NOUN
bracis-19021	170	6	for	for	ADP
bracis-19021	170	7	both	both	PRON
bracis-19021	170	8	estimators.footnote	estimators.footnote	NUM
bracis-19021	170	9	6	6	NUM
bracis-19021	170	10	the	the	DET
bracis-19021	170	11	results	result	NOUN
bracis-19021	170	12	in	in	ADP
bracis-19021	170	13	fig	fig	NOUN
bracis-19021	170	14	.	.	PUNCT
bracis-19021	170	15	 	 	SPACE
bracis-19021	170	16	6	6	NUM
bracis-19021	170	17	suggest	suggest	VERB
bracis-19021	170	18	that	that	SCONJ
bracis-19021	170	19	the	the	DET
bracis-19021	170	20	better	well	ADJ
bracis-19021	170	21	quality	quality	NOUN
bracis-19021	170	22	metrics	metric	NOUN
bracis-19021	170	23	observed	observe	VERB
bracis-19021	170	24	for	for	ADP
bracis-19021	170	25	maac(k	maac(k	NOUN
bracis-19021	170	26	)	)	PUNCT
bracis-19021	170	27	in	in	ADP
bracis-19021	170	28	figs	fig	NOUN
bracis-19021	170	29	.	.	PUNCT
bracis-19021	170	30	 	 	SPACE
bracis-19021	171	1	3	3	NUM
bracis-19021	171	2	and	and	CCONJ
bracis-19021	171	3	4	4	NUM
bracis-19021	171	4	do	do	AUX
bracis-19021	171	5	translate	translate	VERB
bracis-19021	171	6	to	to	ADP
bracis-19021	171	7	faster	fast	ADJ
bracis-19021	171	8	and	and	CCONJ
bracis-19021	171	9	more	more	ADV
bracis-19021	171	10	stable	stable	ADJ
bracis-19021	171	11	policy	policy	NOUN
bracis-19021	171	12	optimization	optimization	NOUN
bracis-19021	171	13	.	.	PUNCT
bracis-19021	172	1	however	however	ADV
bracis-19021	172	2	,	,	PUNCT
bracis-19021	172	3	if	if	SCONJ
bracis-19021	172	4	we	we	PRON
bracis-19021	172	5	normalize	normalize	VERB
bracis-19021	172	6	the	the	DET
bracis-19021	172	7	gradient	gradient	NOUN
bracis-19021	172	8	estimates	estimate	NOUN
bracis-19021	172	9	before	before	ADP
bracis-19021	172	10	passing	pass	VERB
bracis-19021	172	11	them	they	PRON
bracis-19021	172	12	to	to	ADP
bracis-19021	172	13	sgd	sgd	PROPN
bracis-19021	172	14	,	,	PUNCT
bracis-19021	172	15	as	as	ADP
bracis-19021	172	16	in	in	ADP
bracis-19021	172	17	fig	fig	NOUN
bracis-19021	172	18	.	.	PUNCT
bracis-19021	172	19	 	 	SPACE
bracis-19021	172	20	7	7	NUM
bracis-19021	172	21	,	,	PUNCT
bracis-19021	172	22	we	we	PRON
bracis-19021	172	23	see	see	VERB
bracis-19021	172	24	that	that	SCONJ
bracis-19021	172	25	both	both	DET
bracis-19021	172	26	estimators	estimator	NOUN
bracis-19021	172	27	are	be	AUX
bracis-19021	172	28	evenly	evenly	ADV
bracis-19021	172	29	matched	match	VERB
bracis-19021	172	30	.	.	PUNCT
bracis-19021	173	1	these	these	DET
bracis-19021	173	2	results	result	NOUN
bracis-19021	173	3	suggest	suggest	VERB
bracis-19021	173	4	that	that	SCONJ
bracis-19021	173	5	the	the	DET
bracis-19021	173	6	main	main	ADJ
bracis-19021	173	7	advantage	advantage	NOUN
bracis-19021	173	8	of	of	ADP
bracis-19021	173	9	maac(k	maac(k	NOUN
bracis-19021	173	10	)	)	PUNCT
bracis-19021	173	11	over	over	ADP
bracis-19021	173	12	dpg(k	dpg(k	NOUN
bracis-19021	173	13	)	)	PUNCT
bracis-19021	173	14	is	be	AUX
bracis-19021	173	15	in	in	ADP
bracis-19021	173	16	its	its	PRON
bracis-19021	173	17	stronger	strong	ADJ
bracis-19021	173	18	gradient	gradient	NOUN
bracis-19021	173	19	norm	norm	NOUN
bracis-19021	173	20	(	(	PUNCT
bracis-19021	173	21	see	see	VERB
bracis-19021	173	22	fig	fig	NOUN
bracis-19021	173	23	.	.	PUNCT
bracis-19021	173	24	 	 	SPACE
bracis-19021	173	25	5	5	NUM
bracis-19021	173	26	)	)	PUNCT
bracis-19021	173	27	,	,	PUNCT
bracis-19021	173	28	which	which	PRON
bracis-19021	173	29	has	have	AUX
bracis-19021	173	30	been	be	AUX
bracis-19021	173	31	alluded	allude	VERB
bracis-19021	173	32	to	to	ADP
bracis-19021	173	33	in	in	ADP
bracis-19021	173	34	previous	previous	ADJ
bracis-19021	173	35	work	work	NOUN
bracis-19021	173	36	as	as	ADP
bracis-19021	173	37	a	a	DET
bracis-19021	173	38	“	"	PUNCT
bracis-19021	173	39	strong	strong	ADJ
bracis-19021	173	40	learning	learning	NOUN
bracis-19021	173	41	signal	signal	NOUN
bracis-19021	173	42	”	"	PUNCT
bracis-19021	174	1	[	[	X
bracis-19021	174	2	4	4	NUM
bracis-19021	174	3	]	]	PUNCT
bracis-19021	174	4	,	,	PUNCT
bracis-19021	174	5	inducing	induce	VERB
bracis-19021	174	6	a	a	DET
bracis-19021	174	7	faster	fast	ADJ
bracis-19021	174	8	learning	learning	NOUN
bracis-19021	174	9	rate	rate	NOUN
bracis-19021	174	10	.	.	PUNCT
bracis-19021	175	1	table	table	NOUN
bracis-19021	175	2	1	1	NUM
bracis-19021	175	3	.	.	PUNCT
bracis-19021	176	1	median	median	ADJ
bracis-19021	176	2	suboptimality	suboptimality	PROPN
bracis-19021	176	3	gap	gap	NOUN
bracis-19021	176	4	,	,	PUNCT
bracis-19021	176	5	the	the	DET
bracis-19021	176	6	percentage	percentage	NOUN
bracis-19021	176	7	difference	difference	NOUN
bracis-19021	176	8	in	in	ADP
bracis-19021	176	9	expected	expect	VERB
bracis-19021	176	10	return	return	NOUN
bracis-19021	176	11	against	against	ADP
bracis-19021	176	12	the	the	DET
bracis-19021	176	13	optimal	optimal	ADJ
bracis-19021	176	14	policy	policy	NOUN
bracis-19021	176	15	,	,	PUNCT
bracis-19021	176	16	across	across	ADP
bracis-19021	176	17	10	10	NUM
bracis-19021	176	18	seeds	seed	NOUN
bracis-19021	176	19	.	.	PUNCT
bracis-19021	177	1	lqg	lqg	PROPN
bracis-19021	177	2	dimension	dimension	NOUN
bracis-19021	177	3	refers	refer	VERB
bracis-19021	177	4	to	to	ADP
bracis-19021	177	5	the	the	DET
bracis-19021	177	6	dimension	dimension	NOUN
bracis-19021	177	7	of	of	ADP
bracis-19021	177	8	state	state	NOUN
bracis-19021	177	9	and	and	CCONJ
bracis-19021	177	10	action	action	NOUN
bracis-19021	177	11	spaces	space	NOUN
bracis-19021	177	12	.	.	PUNCT
bracis-19021	178	1	we	we	PRON
bracis-19021	178	2	use	use	VERB
bracis-19021	178	3	\(k	\(k	NOUN
bracis-19021	178	4	=	=	SYM
bracis-19021	178	5	8\	8\	NUM
bracis-19021	178	6	)	)	PUNCT
bracis-19021	178	7	and	and	CCONJ
bracis-19021	178	8	\(b=20\	\(b=20\	NOUN
bracis-19021	178	9	)	)	PUNCT
bracis-19021	178	10	for	for	ADP
bracis-19021	178	11	both	both	PRON
bracis-19021	178	12	estimators.full	estimators.full	NUM
bracis-19021	178	13	size	size	NOUN
bracis-19021	178	14	table	table	NOUN
bracis-19021	178	15	fig	fig	NOUN
bracis-19021	178	16	.	.	PUNCT
bracis-19021	179	1	4	4	X
bracis-19021	179	2	.	.	X
bracis-19021	179	3	gradient	gradient	ADJ
bracis-19021	179	4	precision	precision	NOUN
bracis-19021	179	5	for	for	ADP
bracis-19021	179	6	each	each	DET
bracis-19021	179	7	estimator	estimator	NOUN
bracis-19021	179	8	for	for	ADP
bracis-19021	179	9	different	different	ADJ
bracis-19021	179	10	minibatch	minibatch	NOUN
bracis-19021	179	11	sizes	size	NOUN
bracis-19021	179	12	(	(	PUNCT
bracis-19021	179	13	\(b	\(b	NOUN
bracis-19021	179	14	=	=	SYM
bracis-19021	179	15	\	\	NOUN
bracis-19021	179	16	)	)	PUNCT
bracis-19021	179	17	#	#	NOUN
bracis-19021	179	18	samples	sample	NOUN
bracis-19021	179	19	)	)	PUNCT
bracis-19021	179	20	and	and	CCONJ
bracis-19021	179	21	virtual	virtual	ADJ
bracis-19021	179	22	rollout	rollout	NOUN
bracis-19021	179	23	lengths	length	NOUN
bracis-19021	179	24	(	(	PUNCT
bracis-19021	179	25	k	k	NOUN
bracis-19021	179	26	)	)	PUNCT
bracis-19021	179	27	.	.	PUNCT
bracis-19021	180	1	full	full	ADJ
bracis-19021	180	2	size	size	NOUN
bracis-19021	180	3	image	image	NOUN
bracis-19021	180	4	fig	fig	NOUN
bracis-19021	180	5	.	.	PUNCT
bracis-19021	181	1	5	5	NUM
bracis-19021	181	2	.	.	X
bracis-19021	181	3	gradient	gradient	ADJ
bracis-19021	181	4	norm	norm	NOUN
bracis-19021	181	5	for	for	ADP
bracis-19021	181	6	each	each	DET
bracis-19021	181	7	estimator	estimator	NOUN
bracis-19021	181	8	for	for	ADP
bracis-19021	181	9	different	different	ADJ
bracis-19021	181	10	minibatch	minibatch	NOUN
bracis-19021	181	11	sizes	size	NOUN
bracis-19021	181	12	(	(	PUNCT
bracis-19021	181	13	\(b	\(b	NOUN
bracis-19021	181	14	=	=	SYM
bracis-19021	181	15	\	\	NOUN
bracis-19021	181	16	)	)	PUNCT
bracis-19021	181	17	#	#	NOUN
bracis-19021	181	18	samples	sample	NOUN
bracis-19021	181	19	)	)	PUNCT
bracis-19021	181	20	and	and	CCONJ
bracis-19021	181	21	virtual	virtual	ADJ
bracis-19021	181	22	rollout	rollout	NOUN
bracis-19021	181	23	lengths	length	NOUN
bracis-19021	181	24	(	(	PUNCT
bracis-19021	181	25	k	k	NOUN
bracis-19021	181	26	)	)	PUNCT
bracis-19021	181	27	.	.	PUNCT
bracis-19021	182	1	full	full	ADJ
bracis-19021	182	2	size	size	NOUN
bracis-19021	182	3	image	image	NOUN
bracis-19021	182	4	fig	fig	NOUN
bracis-19021	182	5	.	.	PUNCT
bracis-19021	183	1	6	6	NUM
bracis-19021	183	2	.	.	X
bracis-19021	183	3	policy	policy	NOUN
bracis-19021	183	4	optimization	optimization	NOUN
bracis-19021	183	5	with	with	ADP
bracis-19021	183	6	unnormalized	unnormalize	VERB
bracis-19021	183	7	svg	svg	NOUN
bracis-19021	183	8	estimation	estimation	NOUN
bracis-19021	183	9	.	.	PUNCT
bracis-19021	184	1	each	each	DET
bracis-19021	184	2	panel	panel	NOUN
bracis-19021	184	3	corresponds	correspond	VERB
bracis-19021	184	4	to	to	ADP
bracis-19021	184	5	a	a	DET
bracis-19021	184	6	different	different	ADJ
bracis-19021	184	7	lqg	lqg	NOUN
bracis-19021	184	8	instance	instance	NOUN
bracis-19021	184	9	(	(	PUNCT
bracis-19021	184	10	generated	generate	VERB
bracis-19021	184	11	via	via	ADP
bracis-19021	184	12	different	different	ADJ
bracis-19021	184	13	random	random	ADJ
bracis-19021	184	14	seeds	seed	NOUN
bracis-19021	184	15	)	)	PUNCT
bracis-19021	184	16	.	.	PUNCT
bracis-19021	185	1	lines	line	NOUN
bracis-19021	185	2	denote	denote	VERB
bracis-19021	185	3	the	the	DET
bracis-19021	185	4	average	average	ADJ
bracis-19021	185	5	results	result	NOUN
bracis-19021	185	6	and	and	CCONJ
bracis-19021	185	7	shaded	shaded	ADJ
bracis-19021	185	8	regions	region	NOUN
bracis-19021	185	9	,	,	PUNCT
bracis-19021	185	10	one	one	NUM
bracis-19021	185	11	standard	standard	ADJ
bracis-19021	185	12	deviation	deviation	NOUN
bracis-19021	185	13	,	,	PUNCT
bracis-19021	185	14	across	across	ADP
bracis-19021	185	15	10	10	NUM
bracis-19021	185	16	runs	run	NOUN
bracis-19021	185	17	of	of	ADP
bracis-19021	185	18	the	the	DET
bracis-19021	185	19	algorithm	algorithm	NOUN
bracis-19021	185	20	,	,	PUNCT
bracis-19021	185	21	each	each	PRON
bracis-19021	185	22	with	with	ADP
bracis-19021	185	23	a	a	DET
bracis-19021	185	24	different	different	ADJ
bracis-19021	185	25	random	random	ADJ
bracis-19021	185	26	initial	initial	ADJ
bracis-19021	185	27	policy	policy	NOUN
bracis-19021	185	28	.	.	PUNCT
bracis-19021	186	1	results	result	NOUN
bracis-19021	186	2	obtained	obtain	VERB
bracis-19021	186	3	with	with	ADP
bracis-19021	186	4	the	the	DET
bracis-19021	186	5	8	8	NUM
bracis-19021	186	6	-	-	PUNCT
bracis-19021	186	7	step	step	NOUN
bracis-19021	186	8	versions	version	NOUN
bracis-19021	186	9	of	of	ADP
bracis-19021	186	10	each	each	DET
bracis-19021	186	11	estimator	estimator	NOUN
bracis-19021	186	12	.	.	PUNCT
bracis-19021	187	1	full	full	ADJ
bracis-19021	187	2	size	size	NOUN
bracis-19021	187	3	image	image	NOUN
bracis-19021	187	4	fig	fig	NOUN
bracis-19021	187	5	.	.	PUNCT
bracis-19021	188	1	7	7	X
bracis-19021	188	2	.	.	X
bracis-19021	188	3	policy	policy	NOUN
bracis-19021	188	4	optimization	optimization	NOUN
bracis-19021	188	5	with	with	ADP
bracis-19021	188	6	normalized	normalize	VERB
bracis-19021	188	7	svg	svg	PROPN
bracis-19021	188	8	estimation	estimation	NOUN
bracis-19021	188	9	.	.	PUNCT
bracis-19021	189	1	each	each	DET
bracis-19021	189	2	panel	panel	NOUN
bracis-19021	189	3	corresponds	correspond	VERB
bracis-19021	189	4	to	to	ADP
bracis-19021	189	5	a	a	DET
bracis-19021	189	6	different	different	ADJ
bracis-19021	189	7	lqg	lqg	NOUN
bracis-19021	189	8	instance	instance	NOUN
bracis-19021	189	9	(	(	PUNCT
bracis-19021	189	10	generated	generate	VERB
bracis-19021	189	11	via	via	ADP
bracis-19021	189	12	different	different	ADJ
bracis-19021	189	13	random	random	ADJ
bracis-19021	189	14	seeds	seed	NOUN
bracis-19021	189	15	)	)	PUNCT
bracis-19021	189	16	.	.	PUNCT
bracis-19021	190	1	lines	line	NOUN
bracis-19021	190	2	denote	denote	VERB
bracis-19021	190	3	the	the	DET
bracis-19021	190	4	average	average	ADJ
bracis-19021	190	5	results	result	NOUN
bracis-19021	190	6	and	and	CCONJ
bracis-19021	190	7	shaded	shaded	ADJ
bracis-19021	190	8	regions	region	NOUN
bracis-19021	190	9	,	,	PUNCT
bracis-19021	190	10	one	one	NUM
bracis-19021	190	11	standard	standard	ADJ
bracis-19021	190	12	deviation	deviation	NOUN
bracis-19021	190	13	,	,	PUNCT
bracis-19021	190	14	across	across	ADP
bracis-19021	190	15	10	10	NUM
bracis-19021	190	16	runs	run	NOUN
bracis-19021	190	17	of	of	ADP
bracis-19021	190	18	the	the	DET
bracis-19021	190	19	algorithm	algorithm	NOUN
bracis-19021	190	20	,	,	PUNCT
bracis-19021	190	21	each	each	PRON
bracis-19021	190	22	with	with	ADP
bracis-19021	190	23	a	a	DET
bracis-19021	190	24	different	different	ADJ
bracis-19021	190	25	random	random	ADJ
bracis-19021	190	26	initial	initial	ADJ
bracis-19021	190	27	policy	policy	NOUN
bracis-19021	190	28	.	.	PUNCT
bracis-19021	191	1	results	result	NOUN
bracis-19021	191	2	obtained	obtain	VERB
bracis-19021	191	3	with	with	ADP
bracis-19021	191	4	the	the	DET
bracis-19021	191	5	8	8	NUM
bracis-19021	191	6	-	-	PUNCT
bracis-19021	191	7	step	step	NOUN
bracis-19021	191	8	versions	version	NOUN
bracis-19021	191	9	of	of	ADP
bracis-19021	191	10	each	each	DET
bracis-19021	191	11	estimator	estimator	NOUN
bracis-19021	191	12	.	.	PUNCT
bracis-19021	192	1	gradients	gradient	NOUN
bracis-19021	192	2	were	be	AUX
bracis-19021	192	3	normalized	normalize	VERB
bracis-19021	192	4	before	before	ADP
bracis-19021	192	5	being	be	AUX
bracis-19021	192	6	passed	pass	VERB
bracis-19021	192	7	to	to	ADP
bracis-19021	192	8	sgd	sgd	PROPN
bracis-19021	192	9	.	.	PUNCT
bracis-19021	193	1	full	full	ADJ
bracis-19021	193	2	size	size	NOUN
bracis-19021	193	3	image	image	NOUN
bracis-19021	193	4	we	we	PRON
bracis-19021	193	5	also	also	ADV
bracis-19021	193	6	evaluate	evaluate	VERB
bracis-19021	193	7	if	if	SCONJ
bracis-19021	193	8	our	our	PRON
bracis-19021	193	9	previous	previous	ADJ
bracis-19021	193	10	findings	finding	NOUN
bracis-19021	193	11	generalize	generalize	VERB
bracis-19021	193	12	to	to	ADP
bracis-19021	193	13	higher	high	ADJ
bracis-19021	193	14	state	state	NOUN
bracis-19021	193	15	-	-	PUNCT
bracis-19021	193	16	action	action	NOUN
bracis-19021	193	17	space	space	NOUN
bracis-19021	193	18	dimensions	dimension	NOUN
bracis-19021	193	19	,	,	PUNCT
bracis-19021	193	20	where	where	SCONJ
bracis-19021	193	21	sample	sample	NOUN
bracis-19021	193	22	-	-	PUNCT
bracis-19021	193	23	based	base	VERB
bracis-19021	193	24	estimation	estimation	NOUN
bracis-19021	193	25	gets	get	VERB
bracis-19021	193	26	progressively	progressively	ADV
bracis-19021	193	27	harder	hard	ADJ
bracis-19021	193	28	.	.	PUNCT
bracis-19021	194	1	our	our	PRON
bracis-19021	194	2	performance	performance	NOUN
bracis-19021	194	3	metric	metric	ADJ
bracis-19021	194	4	is	be	AUX
bracis-19021	194	5	the	the	DET
bracis-19021	194	6	suboptimality	suboptimality	NOUN
bracis-19021	194	7	gap	gap	NOUN
bracis-19021	194	8	,	,	PUNCT
bracis-19021	194	9	i.e.	i.e.	X
bracis-19021	194	10	,	,	PUNCT
bracis-19021	194	11	the	the	DET
bracis-19021	194	12	percentage	percentage	NOUN
bracis-19021	194	13	difference	difference	NOUN
bracis-19021	194	14	in	in	ADP
bracis-19021	194	15	expected	expected	ADJ
bracis-19021	194	16	return	return	NOUN
bracis-19021	194	17	between	between	ADP
bracis-19021	194	18	the	the	DET
bracis-19021	194	19	current	current	ADJ
bracis-19021	194	20	policy	policy	NOUN
bracis-19021	194	21	and	and	CCONJ
bracis-19021	194	22	the	the	DET
bracis-19021	194	23	optimal	optimal	ADJ
bracis-19021	194	24	one	one	NOUN
bracis-19021	194	25	:	:	PUNCT
bracis-19021	194	26	\(100	\(100	ADV
bracis-19021	194	27	\times	\time	VERB
bracis-19021	194	28	(	(	PUNCT
bracis-19021	194	29	j({\mu	j({\mu	NOUN
bracis-19021	194	30	}	}	PUNCT
bracis-19021	194	31	_	_	PUNCT
bracis-19021	194	32	{	{	PUNCT
bracis-19021	194	33	{	{	PUNCT
bracis-19021	194	34	\theta	\theta	PROPN
bracis-19021	194	35	}	}	PUNCT
bracis-19021	194	36	}	}	PUNCT
bracis-19021	194	37	^\star	^\star	NUM
bracis-19021	194	38	)	)	PUNCT
bracis-19021	194	39	j({\mu	j({\mu	NOUN
bracis-19021	194	40	}	}	PUNCT
bracis-19021	195	1	_	_	PUNCT
bracis-19021	195	2	{	{	PUNCT
bracis-19021	195	3	{	{	PUNCT
bracis-19021	195	4	\theta	\theta	PROPN
bracis-19021	195	5	}	}	PUNCT
bracis-19021	195	6	}	}	PUNCT
bracis-19021	195	7	)	)	PUNCT
bracis-19021	195	8	)	)	PUNCT
bracis-19021	195	9	/j({\mu	/j({\mu	PUNCT
bracis-19021	196	1	}	}	PUNCT
bracis-19021	196	2	_	_	PUNCT
bracis-19021	196	3	{	{	PUNCT
bracis-19021	196	4	{	{	PUNCT
bracis-19021	196	5	\theta	\theta	PROPN
bracis-19021	196	6	}	}	PUNCT
bracis-19021	196	7	}	}	PUNCT
bracis-19021	196	8	^\star	^\star	NUM
bracis-19021	196	9	)	)	PUNCT
bracis-19021	196	10	\).footnote	\).footnote	PROPN
bracis-19021	196	11	7	7	NUM
bracis-19021	196	12	table	table	NOUN
bracis-19021	196	13	 	 	SPACE
bracis-19021	196	14	1	1	NUM
bracis-19021	196	15	summarizes	summarize	VERB
bracis-19021	196	16	our	our	PRON
bracis-19021	196	17	results	result	NOUN
bracis-19021	196	18	with	with	ADP
bracis-19021	196	19	policy	policy	NOUN
bracis-19021	196	20	optimization	optimization	NOUN
bracis-19021	196	21	with	with	ADP
bracis-19021	196	22	varying	vary	VERB
bracis-19021	196	23	lqg	lqg	NOUN
bracis-19021	196	24	sizes	size	NOUN
bracis-19021	196	25	and	and	CCONJ
bracis-19021	196	26	time	time	NOUN
bracis-19021	196	27	budgets.footnote	budgets.footnote	NUM
bracis-19021	196	28	8	8	NUM
bracis-19021	196	29	we	we	PRON
bracis-19021	196	30	do	do	AUX
bracis-19021	196	31	n’t	not	PART
bracis-19021	196	32	normalize	normalize	VERB
bracis-19021	196	33	gradients	gradient	NOUN
bracis-19021	196	34	in	in	ADP
bracis-19021	196	35	this	this	DET
bracis-19021	196	36	case	case	NOUN
bracis-19021	196	37	,	,	PUNCT
bracis-19021	196	38	as	as	SCONJ
bracis-19021	196	39	that	that	PRON
bracis-19021	196	40	is	be	AUX
bracis-19021	196	41	not	not	PART
bracis-19021	196	42	a	a	DET
bracis-19021	196	43	common	common	ADJ
bracis-19021	196	44	practice	practice	NOUN
bracis-19021	196	45	in	in	ADP
bracis-19021	196	46	svg	svg	PROPN
bracis-19021	196	47	algorithms.footnote	algorithms.footnote	PROPN
bracis-19021	196	48	9	9	NUM
bracis-19021	196	49	our	our	PRON
bracis-19021	196	50	findings	finding	NOUN
bracis-19021	196	51	show	show	VERB
bracis-19021	196	52	that	that	SCONJ
bracis-19021	196	53	the	the	DET
bracis-19021	196	54	performance	performance	NOUN
bracis-19021	196	55	gap	gap	NOUN
bracis-19021	196	56	between	between	ADP
bracis-19021	196	57	dpg(k	dpg(k	NOUN
bracis-19021	196	58	)	)	PUNCT
bracis-19021	196	59	and	and	CCONJ
bracis-19021	196	60	maac(k	maac(k	NOUN
bracis-19021	196	61	)	)	PUNCT
bracis-19021	196	62	tends	tend	VERB
bracis-19021	196	63	to	to	PART
bracis-19021	196	64	widen	widen	VERB
bracis-19021	196	65	with	with	ADP
bracis-19021	196	66	higher	high	ADJ
bracis-19021	196	67	dimensionalities	dimensionality	NOUN
bracis-19021	196	68	,	,	PUNCT
bracis-19021	196	69	with	with	ADP
bracis-19021	196	70	policies	policy	NOUN
bracis-19021	196	71	trained	train	VERB
bracis-19021	196	72	via	via	ADP
bracis-19021	196	73	the	the	DET
bracis-19021	196	74	latter	latter	ADJ
bracis-19021	196	75	outperforming	outperforming	NOUN
bracis-19021	196	76	those	those	PRON
bracis-19021	196	77	using	use	VERB
bracis-19021	196	78	the	the	DET
bracis-19021	196	79	former	former	ADJ
bracis-19021	196	80	.	.	PUNCT
bracis-19021	197	1	these	these	DET
bracis-19021	197	2	results	result	NOUN
bracis-19021	197	3	further	far	ADV
bracis-19021	197	4	emphasize	emphasize	VERB
bracis-19021	197	5	the	the	DET
bracis-19021	197	6	practicality	practicality	NOUN
bracis-19021	197	7	of	of	ADP
bracis-19021	197	8	maac(k	maac(k	NOUN
bracis-19021	197	9	)	)	PUNCT
bracis-19021	197	10	over	over	ADP
bracis-19021	197	11	dpg(k	dpg(k	NOUN
bracis-19021	197	12	)	)	PUNCT
bracis-19021	197	13	,	,	PUNCT
bracis-19021	197	14	justifying	justify	VERB
bracis-19021	197	15	the	the	DET
bracis-19021	197	16	former	former	ADJ
bracis-19021	197	17	’s	’s	PART
bracis-19021	197	18	use	use	NOUN
bracis-19021	197	19	in	in	ADP
bracis-19021	197	20	recent	recent	ADJ
bracis-19021	197	21	svg	svg	NOUN
bracis-19021	197	22	methods	method	NOUN
bracis-19021	197	23	[	[	X
bracis-19021	197	24	1	1	NUM
bracis-19021	197	25	,	,	PUNCT
bracis-19021	197	26	4	4	NUM
bracis-19021	197	27	,	,	PUNCT
bracis-19021	197	28	8	8	NUM
bracis-19021	197	29	]	]	PUNCT
bracis-19021	197	30	.	.	PUNCT
bracis-19021	198	1	6	6	NUM
bracis-19021	198	2	conclusions	conclusion	NOUN
bracis-19021	198	3	and	and	CCONJ
bracis-19021	198	4	future	future	ADJ
bracis-19021	198	5	work	work	NOUN
bracis-19021	198	6	in	in	ADP
bracis-19021	198	7	this	this	DET
bracis-19021	198	8	work	work	NOUN
bracis-19021	198	9	,	,	PUNCT
bracis-19021	198	10	we	we	PRON
bracis-19021	198	11	take	take	VERB
bracis-19021	198	12	an	an	DET
bracis-19021	198	13	important	important	ADJ
bracis-19021	198	14	step	step	NOUN
bracis-19021	198	15	towards	towards	ADP
bracis-19021	198	16	a	a	DET
bracis-19021	198	17	better	well	ADJ
bracis-19021	198	18	understanding	understanding	NOUN
bracis-19021	198	19	of	of	ADP
bracis-19021	198	20	current	current	ADJ
bracis-19021	198	21	svg	svg	NOUN
bracis-19021	198	22	methods	method	NOUN
bracis-19021	198	23	.	.	PUNCT
bracis-19021	199	1	using	use	VERB
bracis-19021	199	2	the	the	DET
bracis-19021	199	3	lqg	lqg	NOUN
bracis-19021	199	4	framework	framework	NOUN
bracis-19021	199	5	,	,	PUNCT
bracis-19021	199	6	we	we	PRON
bracis-19021	199	7	show	show	VERB
bracis-19021	199	8	that	that	SCONJ
bracis-19021	199	9	the	the	DET
bracis-19021	199	10	gradient	gradient	NOUN
bracis-19021	199	11	estimation	estimation	NOUN
bracis-19021	199	12	used	use	VERB
bracis-19021	199	13	by	by	ADP
bracis-19021	199	14	maac	maac	NOUN
bracis-19021	199	15	and	and	CCONJ
bracis-19021	199	16	similar	similar	ADJ
bracis-19021	199	17	methods	method	NOUN
bracis-19021	199	18	induces	induce	VERB
bracis-19021	199	19	a	a	DET
bracis-19021	199	20	slight	slight	ADJ
bracis-19021	199	21	bias	bias	NOUN
bracis-19021	199	22	compared	compare	VERB
bracis-19021	199	23	to	to	ADP
bracis-19021	199	24	the	the	DET
bracis-19021	199	25	true	true	ADJ
bracis-19021	199	26	value	value	NOUN
bracis-19021	199	27	gradient	gradient	NOUN
bracis-19021	199	28	.	.	PUNCT
bracis-19021	200	1	on	on	ADP
bracis-19021	200	2	the	the	DET
bracis-19021	200	3	other	other	ADJ
bracis-19021	200	4	hand	hand	NOUN
bracis-19021	200	5	,	,	PUNCT
bracis-19021	200	6	using	use	VERB
bracis-19021	200	7	a	a	DET
bracis-19021	200	8	corresponding	corresponding	ADJ
bracis-19021	200	9	unbiased	unbiased	ADJ
bracis-19021	200	10	estimator	estimator	NOUN
bracis-19021	200	11	such	such	ADJ
bracis-19021	200	12	as	as	ADP
bracis-19021	200	13	the	the	DET
bracis-19021	200	14	k	k	ADJ
bracis-19021	200	15	-	-	PUNCT
bracis-19021	200	16	step	step	NOUN
bracis-19021	200	17	dpg	dpg	PROPN
bracis-19021	200	18	one	one	NUM
bracis-19021	200	19	increases	increase	VERB
bracis-19021	200	20	sample	sample	NOUN
bracis-19021	200	21	-	-	PUNCT
bracis-19021	200	22	complexity	complexity	NOUN
bracis-19021	200	23	due	due	ADP
bracis-19021	200	24	to	to	ADP
bracis-19021	200	25	high	high	ADJ
bracis-19021	200	26	variance	variance	NOUN
bracis-19021	200	27	.	.	PUNCT
bracis-19021	201	1	moreover	moreover	ADV
bracis-19021	201	2	,	,	PUNCT
bracis-19021	201	3	the	the	DET
bracis-19021	201	4	maac	maac	PROPN
bracis-19021	201	5	gradient	gradient	NOUN
bracis-19021	201	6	estimates	estimate	NOUN
bracis-19021	201	7	have	have	VERB
bracis-19021	201	8	higher	high	ADJ
bracis-19021	201	9	magnitudes	magnitude	NOUN
bracis-19021	201	10	,	,	PUNCT
bracis-19021	201	11	which	which	PRON
bracis-19021	201	12	could	could	AUX
bracis-19021	201	13	help	help	VERB
bracis-19021	201	14	explain	explain	VERB
bracis-19021	201	15	the	the	DET
bracis-19021	201	16	fast	fast	ADJ
bracis-19021	201	17	learning	learning	NOUN
bracis-19021	201	18	performance	performance	NOUN
bracis-19021	201	19	of	of	ADP
bracis-19021	201	20	current	current	ADJ
bracis-19021	201	21	methods	method	NOUN
bracis-19021	201	22	.	.	PUNCT
bracis-19021	202	1	indeed	indeed	ADV
bracis-19021	202	2	,	,	PUNCT
bracis-19021	202	3	we	we	PRON
bracis-19021	202	4	found	find	VERB
bracis-19021	202	5	that	that	SCONJ
bracis-19021	202	6	policies	policy	NOUN
bracis-19021	202	7	trained	train	VERB
bracis-19021	202	8	with	with	ADP
bracis-19021	202	9	maac	maac	PROPN
bracis-19021	202	10	converge	converge	VERB
bracis-19021	202	11	faster	fast	ADV
bracis-19021	202	12	to	to	ADP
bracis-19021	202	13	the	the	DET
bracis-19021	202	14	optimal	optimal	ADJ
bracis-19021	202	15	policies	policy	NOUN
bracis-19021	202	16	than	than	ADP
bracis-19021	202	17	those	those	PRON
bracis-19021	202	18	using	use	VERB
bracis-19021	202	19	the	the	DET
bracis-19021	202	20	k	k	ADJ
bracis-19021	202	21	-	-	PUNCT
bracis-19021	202	22	step	step	NOUN
bracis-19021	202	23	dpg	dpg	NOUN
bracis-19021	202	24	across	across	ADP
bracis-19021	202	25	several	several	ADJ
bracis-19021	202	26	lqg	lqg	NOUN
bracis-19021	202	27	instances	instance	NOUN
bracis-19021	202	28	.	.	PUNCT
bracis-19021	203	1	future	future	ADJ
bracis-19021	203	2	work	work	NOUN
bracis-19021	203	3	may	may	AUX
bracis-19021	203	4	further	far	ADV
bracis-19021	203	5	leverage	leverage	VERB
bracis-19021	203	6	the	the	DET
bracis-19021	203	7	lqg	lqg	NOUN
bracis-19021	203	8	framework	framework	NOUN
bracis-19021	203	9	to	to	PART
bracis-19021	203	10	perform	perform	VERB
bracis-19021	203	11	fine	fine	ADV
bracis-19021	203	12	-	-	PUNCT
bracis-19021	203	13	grained	grain	VERB
bracis-19021	203	14	analyses	analysis	NOUN
bracis-19021	203	15	of	of	ADP
bracis-19021	203	16	other	other	ADJ
bracis-19021	203	17	important	important	ADJ
bracis-19021	203	18	components	component	NOUN
bracis-19021	203	19	of	of	ADP
bracis-19021	203	20	svg	svg	PROPN
bracis-19021	203	21	algorithms	algorithm	NOUN
bracis-19021	203	22	.	.	PUNCT
bracis-19021	204	1	for	for	ADP
bracis-19021	204	2	example	example	NOUN
bracis-19021	204	3	,	,	PUNCT
bracis-19021	204	4	little	little	ADJ
bracis-19021	204	5	is	be	AUX
bracis-19021	204	6	known	know	VERB
bracis-19021	204	7	about	about	ADP
bracis-19021	204	8	the	the	DET
bracis-19021	204	9	interplay	interplay	NOUN
bracis-19021	204	10	between	between	ADP
bracis-19021	204	11	model	model	NOUN
bracis-19021	204	12	,	,	PUNCT
bracis-19021	204	13	value	value	NOUN
bracis-19021	204	14	function	function	NOUN
bracis-19021	204	15	,	,	PUNCT
bracis-19021	204	16	and	and	CCONJ
bracis-19021	204	17	policy	policy	NOUN
bracis-19021	204	18	learning	learning	NOUN
bracis-19021	204	19	from	from	ADP
bracis-19021	204	20	data	datum	NOUN
bracis-19021	204	21	in	in	ADP
bracis-19021	204	22	practice	practice	NOUN
bracis-19021	204	23	.	.	PUNCT
bracis-19021	205	1	a	a	DET
bracis-19021	205	2	study	study	NOUN
bracis-19021	205	3	on	on	ADP
bracis-19021	205	4	model	model	NOUN
bracis-19021	205	5	and	and	CCONJ
bracis-19021	205	6	value	value	NOUN
bracis-19021	205	7	function	function	NOUN
bracis-19021	205	8	optimization	optimization	NOUN
bracis-19021	205	9	metrics	metric	NOUN
bracis-19021	205	10	and	and	CCONJ
bracis-19021	205	11	their	their	PRON
bracis-19021	205	12	relation	relation	NOUN
bracis-19021	205	13	to	to	ADP
bracis-19021	205	14	the	the	DET
bracis-19021	205	15	gradient	gradient	NOUN
bracis-19021	205	16	estimation	estimation	NOUN
bracis-19021	205	17	accuracy	accuracy	NOUN
bracis-19021	205	18	and	and	CCONJ
bracis-19021	205	19	precision	precision	NOUN
bracis-19021	205	20	can	can	AUX
bracis-19021	205	21	help	help	VERB
bracis-19021	205	22	in	in	ADP
bracis-19021	205	23	the	the	DET
bracis-19021	205	24	design	design	NOUN
bracis-19021	205	25	of	of	ADP
bracis-19021	205	26	stable	stable	ADJ
bracis-19021	205	27	and	and	CCONJ
bracis-19021	205	28	efficient	efficient	ADJ
bracis-19021	205	29	svg	svg	NOUN
bracis-19021	205	30	algorithms	algorithm	NOUN
bracis-19021	205	31	in	in	ADP
bracis-19021	205	32	the	the	DET
bracis-19021	205	33	future	future	NOUN
bracis-19021	205	34	.	.	PUNCT
bracis-19021	206	1	another	another	DET
bracis-19021	206	2	direction	direction	NOUN
bracis-19021	206	3	for	for	ADP
bracis-19021	206	4	investigation	investigation	NOUN
bracis-19021	206	5	is	be	AUX
bracis-19021	206	6	analyzing	analyze	VERB
bracis-19021	206	7	the	the	DET
bracis-19021	206	8	impact	impact	NOUN
bracis-19021	206	9	of	of	ADP
bracis-19021	206	10	off	off	ADP
bracis-19021	206	11	-	-	PUNCT
bracis-19021	206	12	policy	policy	NOUN
bracis-19021	206	13	data	datum	NOUN
bracis-19021	206	14	collection	collection	NOUN
bracis-19021	206	15	for	for	ADP
bracis-19021	206	16	model	model	NOUN
bracis-19021	206	17	training	training	NOUN
bracis-19021	206	18	.	.	PUNCT
bracis-19021	207	1	since	since	SCONJ
bracis-19021	207	2	models	model	NOUN
bracis-19021	207	3	have	have	VERB
bracis-19021	207	4	limited	limit	VERB
bracis-19021	207	5	representation	representation	NOUN
bracis-19021	207	6	capacity	capacity	NOUN
bracis-19021	207	7	,	,	PUNCT
bracis-19021	207	8	learning	learn	VERB
bracis-19021	207	9	the	the	DET
bracis-19021	207	10	mdp	mdp	NOUN
bracis-19021	207	11	dynamics	dynamic	NOUN
bracis-19021	207	12	from	from	ADP
bracis-19021	207	13	the	the	DET
bracis-19021	207	14	distribution	distribution	NOUN
bracis-19021	207	15	of	of	ADP
bracis-19021	207	16	another	another	DET
bracis-19021	207	17	policy	policy	NOUN
bracis-19021	207	18	may	may	AUX
bracis-19021	207	19	not	not	PART
bracis-19021	207	20	translate	translate	VERB
bracis-19021	207	21	to	to	ADP
bracis-19021	207	22	good	good	ADJ
bracis-19021	207	23	gradient	gradient	NOUN
bracis-19021	207	24	estimation	estimation	NOUN
bracis-19021	207	25	of	of	ADP
bracis-19021	207	26	the	the	DET
bracis-19021	207	27	target	target	NOUN
bracis-19021	207	28	policy	policy	NOUN
bracis-19021	207	29	.	.	PUNCT
bracis-19021	208	1	notes	note	VERB
bracis-19021	208	2	1.our	1.our	PRON
bracis-19021	208	3	formula	formula	NOUN
bracis-19021	208	4	differs	differ	VERB
bracis-19021	208	5	slightly	slightly	ADV
bracis-19021	208	6	from	from	ADP
bracis-19021	208	7	the	the	DET
bracis-19021	208	8	original	original	NOUN
bracis-19021	208	9	in	in	SCONJ
bracis-19021	208	10	that	that	SCONJ
bracis-19021	208	11	it	it	PRON
bracis-19021	208	12	considers	consider	VERB
bracis-19021	208	13	a	a	DET
bracis-19021	208	14	deterministic	deterministic	ADJ
bracis-19021	208	15	policy	policy	NOUN
bracis-19021	208	16	instead	instead	ADV
bracis-19021	208	17	of	of	ADP
bracis-19021	208	18	a	a	DET
bracis-19021	208	19	stochastic	stochastic	ADJ
bracis-19021	208	20	one	one	NUM
bracis-19021	208	21	.	.	PUNCT
bracis-19021	209	1	2.we	2.we	PRON
bracis-19021	209	2	use	use	VERB
bracis-19021	209	3	the	the	DET
bracis-19021	209	4	make_spd_matrix	make_spd_matrix	NOUN
bracis-19021	209	5	function	function	NOUN
bracis-19021	209	6	.	.	PUNCT
bracis-19021	210	1	3.we	3.we	NUM
bracis-19021	210	2	use	use	VERB
bracis-19021	210	3	the	the	DET
bracis-19021	210	4	scipy.signal.place_poles	scipy.signal.place_pole	NOUN
bracis-19021	210	5	function	function	NOUN
bracis-19021	210	6	.	.	PUNCT
bracis-19021	211	1	4.we	4.we	PRON
bracis-19021	211	2	use	use	VERB
bracis-19021	211	3	the	the	DET
bracis-19021	211	4	same	same	ADJ
bracis-19021	211	5	10	10	NUM
bracis-19021	211	6	random	random	ADJ
bracis-19021	211	7	seeds	seed	NOUN
bracis-19021	211	8	for	for	ADP
bracis-19021	211	9	experiments	experiment	NOUN
bracis-19021	211	10	across	across	ADP
bracis-19021	211	11	values	value	NOUN
bracis-19021	211	12	of	of	ADP
bracis-19021	211	13	k.	k.	PROPN
bracis-19021	211	14	5.we	5.we	NUM
bracis-19021	211	15	use	use	VERB
bracis-19021	211	16	seaborn.lineplot	seaborn.lineplot	NOUN
bracis-19021	211	17	to	to	PART
bracis-19021	211	18	produce	produce	VERB
bracis-19021	211	19	the	the	DET
bracis-19021	211	20	aggregated	aggregate	VERB
bracis-19021	211	21	curves	curve	NOUN
bracis-19021	211	22	.	.	PUNCT
bracis-19021	212	1	6.learning	6.learning	NUM
bracis-19021	212	2	rate	rate	NOUN
bracis-19021	212	3	of	of	ADP
bracis-19021	212	4	\(10^{-2}\	\(10^{-2}\	NOUN
bracis-19021	212	5	)	)	PUNCT
bracis-19021	212	6	,	,	PUNCT
bracis-19021	212	7	\(b=200\	\(b=200\	PROPN
bracis-19021	212	8	)	)	PUNCT
bracis-19021	212	9	,	,	PUNCT
bracis-19021	212	10	and	and	CCONJ
bracis-19021	212	11	\(k=8\	\(k=8\	NOUN
bracis-19021	212	12	)	)	PUNCT
bracis-19021	212	13	.	.	PUNCT
bracis-19021	213	1	7.recall	7.recall	NUM
bracis-19021	213	2	from	from	ADP
bracis-19021	213	3	sect	sect	NOUN
bracis-19021	213	4	.	.	PUNCT
bracis-19021	213	5	 	 	SPACE
bracis-19021	214	1	3	3	NUM
bracis-19021	214	2	that	that	PRON
bracis-19021	214	3	lqg	lqg	NOUN
bracis-19021	214	4	allows	allow	VERB
bracis-19021	214	5	us	we	PRON
bracis-19021	214	6	to	to	PART
bracis-19021	214	7	compute	compute	VERB
bracis-19021	214	8	the	the	DET
bracis-19021	214	9	optimal	optimal	ADJ
bracis-19021	214	10	policy	policy	NOUN
bracis-19021	214	11	analytically	analytically	ADV
bracis-19021	214	12	.	.	PUNCT
bracis-19021	215	1	8.we	8.we	NUM
bracis-19021	215	2	found	find	VERB
bracis-19021	215	3	that	that	SCONJ
bracis-19021	215	4	the	the	DET
bracis-19021	215	5	computation	computation	NOUN
bracis-19021	215	6	times	time	NOUN
bracis-19021	215	7	for	for	ADP
bracis-19021	215	8	both	both	DET
bracis-19021	215	9	estimators	estimator	NOUN
bracis-19021	215	10	were	be	AUX
bracis-19021	215	11	equivalent	equivalent	ADJ
bracis-19021	215	12	.	.	PUNCT
bracis-19021	216	1	9.we	9.we	NUM
bracis-19021	216	2	only	only	ADV
bracis-19021	216	3	clip	clip	VERB
bracis-19021	216	4	the	the	DET
bracis-19021	216	5	gradient	gradient	ADJ
bracis-19021	216	6	norm	norm	NOUN
bracis-19021	216	7	at	at	ADP
bracis-19021	216	8	a	a	DET
bracis-19021	216	9	maximum	maximum	NOUN
bracis-19021	216	10	of	of	ADP
bracis-19021	216	11	100	100	NUM
bracis-19021	216	12	to	to	PART
bracis-19021	216	13	avoid	avoid	VERB
bracis-19021	216	14	numerical	numerical	ADJ
bracis-19021	216	15	errors	error	NOUN
bracis-19021	216	16	.	.	PUNCT
bracis-19021	217	1	references	reference	NOUN
bracis-19021	217	2	amos	amos	PROPN
bracis-19021	217	3	,	,	PUNCT
bracis-19021	217	4	b.	b.	PROPN
bracis-19021	217	5	,	,	PUNCT
bracis-19021	217	6	stanton	stanton	PROPN
bracis-19021	217	7	,	,	PUNCT
bracis-19021	217	8	s.	s.	PROPN
bracis-19021	217	9	,	,	PUNCT
bracis-19021	217	10	yarats	yarat	NOUN
bracis-19021	217	11	,	,	PUNCT
bracis-19021	217	12	d.	d.	PROPN
bracis-19021	217	13	,	,	PUNCT
bracis-19021	217	14	wilson	wilson	PROPN
bracis-19021	217	15	,	,	PUNCT
bracis-19021	217	16	a.g	a.g	PROPN
bracis-19021	217	17	.	.	PROPN
bracis-19021	217	18	:	:	PUNCT
bracis-19021	218	1	on	on	ADP
bracis-19021	218	2	the	the	DET
bracis-19021	218	3	model	model	NOUN
bracis-19021	218	4	-	-	PUNCT
bracis-19021	218	5	based	base	VERB
bracis-19021	218	6	stochastic	stochastic	ADJ
bracis-19021	218	7	value	value	NOUN
bracis-19021	218	8	gradient	gradient	NOUN
bracis-19021	218	9	for	for	ADP
bracis-19021	218	10	continuous	continuous	ADJ
bracis-19021	218	11	reinforcement	reinforcement	NOUN
bracis-19021	218	12	learning	learning	NOUN
bracis-19021	218	13	.	.	PUNCT
bracis-19021	219	1	corr	corr	PROPN
bracis-19021	219	2	arxiv:2008.1	arxiv:2008.1	PROPN
bracis-19021	219	3	(	(	PUNCT
bracis-19021	219	4	2020	2020	NUM
bracis-19021	219	5	)	)	PUNCT
bracis-19021	219	6	byravan	byravan	NOUN
bracis-19021	219	7	,	,	PUNCT
bracis-19021	219	8	a.	a.	NOUN
bracis-19021	219	9	,	,	PUNCT
bracis-19021	219	10	et	et	PROPN
bracis-19021	219	11	al	al	PROPN
bracis-19021	219	12	.	.	PROPN
bracis-19021	219	13	:	:	PUNCT
bracis-19021	219	14	imagined	imagine	VERB
bracis-19021	219	15	value	value	NOUN
bracis-19021	219	16	gradients	gradient	NOUN
bracis-19021	219	17	:	:	PUNCT
bracis-19021	219	18	model	model	NOUN
bracis-19021	219	19	-	-	PUNCT
bracis-19021	219	20	based	base	VERB
bracis-19021	219	21	policy	policy	NOUN
bracis-19021	219	22	optimization	optimization	NOUN
bracis-19021	219	23	with	with	ADP
bracis-19021	219	24	transferable	transferable	ADJ
bracis-19021	219	25	latent	latent	NOUN
bracis-19021	219	26	dynamics	dynamic	NOUN
bracis-19021	219	27	models	model	NOUN
bracis-19021	219	28	.	.	PUNCT
bracis-19021	220	1	in	in	ADP
bracis-19021	220	2	:	:	PUNCT
bracis-19021	220	3	corl	corl	PROPN
bracis-19021	220	4	.	.	PUNCT
bracis-19021	220	5	proceedings	proceeding	NOUN
bracis-19021	220	6	of	of	ADP
bracis-19021	220	7	machine	machine	NOUN
bracis-19021	220	8	learning	learn	VERB
bracis-19021	220	9	research	research	NOUN
bracis-19021	220	10	,	,	PUNCT
bracis-19021	220	11	vol	vol	NOUN
bracis-19021	220	12	.	.	PROPN
bracis-19021	220	13	100	100	NUM
bracis-19021	220	14	,	,	PUNCT
bracis-19021	220	15	pp	pp	ADJ
bracis-19021	220	16	.	.	PUNCT
bracis-19021	221	1	566–589	566–589	NUM
bracis-19021	221	2	.	.	PUNCT
bracis-19021	221	3	pmlr	pmlr	NOUN
bracis-19021	221	4	(	(	PUNCT
bracis-19021	221	5	2019	2019	NUM
bracis-19021	221	6	)	)	PUNCT
bracis-19021	221	7	google	google	PROPN
bracis-19021	221	8	scholar	scholar	NOUN
bracis-19021	221	9	  	  	SPACE
bracis-19021	221	10	chan	chan	PROPN
bracis-19021	221	11	,	,	PUNCT
bracis-19021	221	12	s.c.y	s.c.y	PROPN
bracis-19021	221	13	.	.	PROPN
bracis-19021	221	14	,	,	PUNCT
bracis-19021	221	15	fishman	fishman	PROPN
bracis-19021	221	16	,	,	PUNCT
bracis-19021	221	17	s.	s.	PROPN
bracis-19021	221	18	,	,	PUNCT
bracis-19021	221	19	korattikara	korattikara	PROPN
bracis-19021	221	20	,	,	PUNCT
bracis-19021	221	21	a.	a.	NOUN
bracis-19021	221	22	,	,	PUNCT
bracis-19021	221	23	canny	canny	ADJ
bracis-19021	221	24	,	,	PUNCT
bracis-19021	221	25	j.	j.	PROPN
bracis-19021	221	26	,	,	PUNCT
bracis-19021	221	27	guadarrama	guadarrama	NOUN
bracis-19021	221	28	,	,	PUNCT
bracis-19021	221	29	s.	s.	PROPN
bracis-19021	221	30	:	:	PUNCT
bracis-19021	221	31	measuring	measure	VERB
bracis-19021	221	32	the	the	DET
bracis-19021	221	33	reliability	reliability	NOUN
bracis-19021	221	34	of	of	ADP
bracis-19021	221	35	reinforcement	reinforcement	NOUN
bracis-19021	221	36	learning	learning	NOUN
bracis-19021	221	37	algorithms	algorithm	NOUN
bracis-19021	221	38	.	.	PUNCT
bracis-19021	222	1	in	in	ADP
bracis-19021	222	2	:	:	PUNCT
bracis-19021	222	3	iclr	iclr	NOUN
bracis-19021	222	4	.	.	PUNCT
bracis-19021	223	1	openreview.net	openreview.net	X
bracis-19021	223	2	(	(	PUNCT
bracis-19021	223	3	2020	2020	NUM
bracis-19021	223	4	)	)	PUNCT
bracis-19021	223	5	google	google	PROPN
bracis-19021	223	6	scholar	scholar	NOUN
bracis-19021	223	7	  	  	SPACE
bracis-19021	223	8	clavera	clavera	PROPN
bracis-19021	223	9	,	,	PUNCT
bracis-19021	223	10	i.	i.	PROPN
bracis-19021	223	11	,	,	PUNCT
bracis-19021	223	12	fu	fu	PROPN
bracis-19021	223	13	,	,	PUNCT
bracis-19021	223	14	y.	y.	NOUN
bracis-19021	223	15	,	,	PUNCT
bracis-19021	223	16	abbeel	abbeel	NOUN
bracis-19021	223	17	,	,	PUNCT
bracis-19021	223	18	p.	p.	NOUN
bracis-19021	223	19	:	:	PUNCT
bracis-19021	223	20	model	model	ADJ
bracis-19021	223	21	-	-	PUNCT
bracis-19021	223	22	augmented	augment	VERB
bracis-19021	223	23	actor	actor	NOUN
bracis-19021	223	24	-	-	PUNCT
bracis-19021	223	25	critic	critic	NOUN
bracis-19021	223	26	:	:	PUNCT
bracis-19021	223	27	backpropagating	backpropagate	VERB
bracis-19021	223	28	through	through	ADP
bracis-19021	223	29	paths	path	NOUN
bracis-19021	223	30	.	.	PUNCT
bracis-19021	224	1	in	in	ADP
bracis-19021	224	2	:	:	PUNCT
bracis-19021	224	3	iclr	iclr	NOUN
bracis-19021	224	4	.	.	PUNCT
bracis-19021	225	1	openreview.net	openreview.net	X
bracis-19021	225	2	(	(	PUNCT
bracis-19021	225	3	2020	2020	NUM
bracis-19021	225	4	)	)	PUNCT
bracis-19021	225	5	.	.	PUNCT
bracis-19021	226	1	https://openreview.net/forum?id=skln2a4ydb	https://openreview.net/forum?id=skln2a4ydb	PROPN
bracis-19021	226	2	deisenroth	deisenroth	NOUN
bracis-19021	226	3	,	,	PUNCT
bracis-19021	226	4	m.p	m.p	PROPN
bracis-19021	226	5	.	.	PROPN
bracis-19021	226	6	,	,	PUNCT
bracis-19021	226	7	rasmussen	rasmussen	PROPN
bracis-19021	226	8	,	,	PUNCT
bracis-19021	226	9	c.e	c.e	PROPN
bracis-19021	226	10	.	.	PROPN
bracis-19021	226	11	:	:	PUNCT
bracis-19021	227	1	pilco	pilco	NOUN
bracis-19021	227	2	:	:	PUNCT
bracis-19021	227	3	a	a	DET
bracis-19021	227	4	model	model	NOUN
bracis-19021	227	5	-	-	PUNCT
bracis-19021	227	6	based	base	VERB
bracis-19021	227	7	and	and	CCONJ
bracis-19021	227	8	data	data	NOUN
bracis-19021	227	9	-	-	PUNCT
bracis-19021	227	10	efficient	efficient	ADJ
bracis-19021	227	11	approach	approach	NOUN
bracis-19021	227	12	to	to	ADP
bracis-19021	227	13	policy	policy	NOUN
bracis-19021	227	14	search	search	NOUN
bracis-19021	227	15	.	.	PUNCT
bracis-19021	228	1	in	in	ADP
bracis-19021	228	2	:	:	PUNCT
bracis-19021	228	3	getoor	getoor	ADV
bracis-19021	228	4	,	,	PUNCT
bracis-19021	228	5	l.	l.	PROPN
bracis-19021	228	6	,	,	PUNCT
bracis-19021	228	7	scheffer	scheffer	PROPN
bracis-19021	228	8	,	,	PUNCT
bracis-19021	228	9	t.	t.	PROPN
bracis-19021	228	10	(	(	PUNCT
bracis-19021	228	11	eds	eds	PROPN
bracis-19021	228	12	.	.	PUNCT
bracis-19021	228	13	)	)	PUNCT
bracis-19021	228	14	proceedings	proceeding	NOUN
bracis-19021	228	15	of	of	ADP
bracis-19021	228	16	the	the	DET
bracis-19021	228	17	28th	28th	ADJ
bracis-19021	228	18	international	international	ADJ
bracis-19021	228	19	conference	conference	NOUN
bracis-19021	228	20	on	on	ADP
bracis-19021	228	21	machine	machine	NOUN
bracis-19021	228	22	learning	learning	NOUN
bracis-19021	228	23	,	,	PUNCT
bracis-19021	228	24	icml	icml	VERB
bracis-19021	228	25	2011	2011	NUM
bracis-19021	228	26	,	,	PUNCT
bracis-19021	228	27	bellevue	bellevue	PROPN
bracis-19021	228	28	,	,	PUNCT
bracis-19021	228	29	washington	washington	PROPN
bracis-19021	228	30	,	,	PUNCT
bracis-19021	228	31	usa	usa	PROPN
bracis-19021	228	32	,	,	PUNCT
bracis-19021	228	33	28	28	NUM
bracis-19021	228	34	june–2	june–2	PROPN
bracis-19021	228	35	july	july	PROPN
bracis-19021	228	36	2011	2011	NUM
bracis-19021	228	37	,	,	PUNCT
bracis-19021	228	38	pp	pp	ADV
bracis-19021	228	39	.	.	PUNCT
bracis-19021	229	1	465–472	465–472	NUM
bracis-19021	229	2	.	.	PUNCT
bracis-19021	230	1	omnipress	omnipress	ADJ
bracis-19021	230	2	(	(	PUNCT
bracis-19021	230	3	2011	2011	NUM
bracis-19021	230	4	)	)	PUNCT
bracis-19021	230	5	.	.	PUNCT
bracis-19021	231	1	https://icml.cc/2011/papers/323_icmlpaper.pdf	https://icml.cc/2011/papers/323_icmlpaper.pdf	PROPN
bracis-19021	231	2	engstrom	engstrom	PROPN
bracis-19021	231	3	,	,	PUNCT
bracis-19021	231	4	l.	l.	PROPN
bracis-19021	231	5	,	,	PUNCT
bracis-19021	231	6	et	et	PROPN
bracis-19021	231	7	al	al	PROPN
bracis-19021	231	8	.	.	PUNCT
bracis-19021	231	9	:	:	PUNCT
bracis-19021	232	1	implementation	implementation	NOUN
bracis-19021	232	2	matters	matter	VERB
bracis-19021	232	3	in	in	ADP
bracis-19021	232	4	deep	deep	ADJ
bracis-19021	232	5	rl	rl	NOUN
bracis-19021	232	6	:	:	PUNCT
bracis-19021	232	7	a	a	DET
bracis-19021	232	8	case	case	NOUN
bracis-19021	232	9	study	study	NOUN
bracis-19021	232	10	on	on	ADP
bracis-19021	232	11	ppo	ppo	PROPN
bracis-19021	232	12	and	and	CCONJ
bracis-19021	232	13	trpo	trpo	NOUN
bracis-19021	232	14	.	.	PUNCT
bracis-19021	233	1	in	in	ADP
bracis-19021	233	2	:	:	PUNCT
bracis-19021	233	3	iclr	iclr	NOUN
bracis-19021	233	4	.	.	PUNCT
bracis-19021	234	1	openreview.net	openreview.net	X
bracis-19021	234	2	(	(	PUNCT
bracis-19021	234	3	2020	2020	NUM
bracis-19021	234	4	)	)	PUNCT
bracis-19021	234	5	.	.	PUNCT
bracis-19021	235	1	https://github.com/implementation-matters/code-for-paper	https://github.com/implementation-matters/code-for-paper	PUNCT
bracis-19021	235	2	goodfellow	goodfellow	PROPN
bracis-19021	235	3	,	,	PUNCT
bracis-19021	235	4	i.j	i.j	PROPN
bracis-19021	235	5	.	.	PROPN
bracis-19021	235	6	,	,	PUNCT
bracis-19021	235	7	bengio	bengio	PROPN
bracis-19021	235	8	,	,	PUNCT
bracis-19021	235	9	y.	y.	NOUN
bracis-19021	235	10	,	,	PUNCT
bracis-19021	235	11	courville	courville	NOUN
bracis-19021	235	12	,	,	PUNCT
bracis-19021	235	13	a.c	a.c	PROPN
bracis-19021	235	14	.	.	PROPN
bracis-19021	235	15	:	:	PUNCT
bracis-19021	236	1	deep	deep	ADJ
bracis-19021	236	2	learning	learning	NOUN
bracis-19021	236	3	.	.	PUNCT
bracis-19021	237	1	adaptive	adaptive	ADJ
bracis-19021	237	2	computation	computation	NOUN
bracis-19021	237	3	and	and	CCONJ
bracis-19021	237	4	machine	machine	NOUN
bracis-19021	237	5	learning	learning	NOUN
bracis-19021	237	6	.	.	PUNCT
bracis-19021	238	1	mit	mit	PROPN
bracis-19021	238	2	press	press	PROPN
bracis-19021	238	3	,	,	PUNCT
bracis-19021	238	4	cambridge	cambridge	PROPN
bracis-19021	238	5	(	(	PUNCT
bracis-19021	238	6	2016	2016	NUM
bracis-19021	238	7	)	)	PUNCT
bracis-19021	238	8	math	math	NOUN
bracis-19021	238	9	  	  	SPACE
bracis-19021	238	10	google	google	PROPN
bracis-19021	238	11	scholar	scholar	NOUN
bracis-19021	238	12	  	  	SPACE
bracis-19021	238	13	hafner	hafner	PROPN
bracis-19021	238	14	,	,	PUNCT
bracis-19021	238	15	d.	d.	PROPN
bracis-19021	238	16	,	,	PUNCT
bracis-19021	238	17	lillicrap	lillicrap	PROPN
bracis-19021	238	18	,	,	PUNCT
bracis-19021	238	19	t.p	t.p	PROPN
bracis-19021	238	20	.	.	PROPN
bracis-19021	238	21	,	,	PUNCT
bracis-19021	238	22	ba	ba	PROPN
bracis-19021	238	23	,	,	PUNCT
bracis-19021	238	24	j.	j.	PROPN
bracis-19021	238	25	,	,	PUNCT
bracis-19021	238	26	norouzi	norouzi	PROPN
bracis-19021	238	27	,	,	PUNCT
bracis-19021	238	28	m.	m.	NOUN
bracis-19021	238	29	:	:	PUNCT
bracis-19021	238	30	dream	dream	NOUN
bracis-19021	238	31	to	to	PART
bracis-19021	238	32	control	control	VERB
bracis-19021	238	33	:	:	PUNCT
bracis-19021	238	34	learning	learn	VERB
bracis-19021	238	35	behaviors	behavior	NOUN
bracis-19021	238	36	by	by	ADP
bracis-19021	238	37	latent	latent	ADJ
bracis-19021	238	38	imagination	imagination	NOUN
bracis-19021	238	39	.	.	PUNCT
bracis-19021	239	1	in	in	ADP
bracis-19021	239	2	:	:	PUNCT
bracis-19021	239	3	iclr	iclr	NOUN
bracis-19021	239	4	.	.	PUNCT
bracis-19021	240	1	openreview.net	openreview.net	X
bracis-19021	240	2	(	(	PUNCT
bracis-19021	240	3	2020	2020	NUM
bracis-19021	240	4	)	)	PUNCT
bracis-19021	240	5	google	google	PROPN
bracis-19021	240	6	scholar	scholar	NOUN
bracis-19021	240	7	  	  	SPACE
bracis-19021	240	8	heess	heess	PROPN
bracis-19021	240	9	,	,	PUNCT
bracis-19021	240	10	n.	n.	PROPN
bracis-19021	240	11	,	,	PUNCT
bracis-19021	240	12	wayne	wayne	PROPN
bracis-19021	240	13	,	,	PUNCT
bracis-19021	240	14	g.	g.	PROPN
bracis-19021	240	15	,	,	PUNCT
bracis-19021	240	16	silver	silver	NOUN
bracis-19021	240	17	,	,	PUNCT
bracis-19021	240	18	d.	d.	PROPN
bracis-19021	240	19	,	,	PUNCT
bracis-19021	240	20	lillicrap	lillicrap	PROPN
bracis-19021	240	21	,	,	PUNCT
bracis-19021	240	22	t.p	t.p	PROPN
bracis-19021	240	23	.	.	PROPN
bracis-19021	240	24	,	,	PUNCT
bracis-19021	240	25	erez	erez	PROPN
bracis-19021	240	26	,	,	PUNCT
bracis-19021	240	27	t.	t.	PROPN
bracis-19021	240	28	,	,	PUNCT
bracis-19021	240	29	tassa	tassa	PROPN
bracis-19021	240	30	,	,	PUNCT
bracis-19021	240	31	y.	y.	NOUN
bracis-19021	240	32	:	:	PUNCT
bracis-19021	240	33	learning	learn	VERB
bracis-19021	240	34	continuous	continuous	ADJ
bracis-19021	240	35	control	control	NOUN
bracis-19021	240	36	policies	policy	NOUN
bracis-19021	240	37	by	by	ADP
bracis-19021	240	38	stochastic	stochastic	ADJ
bracis-19021	240	39	value	value	NOUN
bracis-19021	240	40	gradients	gradient	NOUN
bracis-19021	240	41	.	.	PUNCT
bracis-19021	241	1	in	in	ADP
bracis-19021	241	2	:	:	PUNCT
bracis-19021	241	3	nips	nip	NOUN
bracis-19021	241	4	,	,	PUNCT
bracis-19021	241	5	pp	pp	ADV
bracis-19021	241	6	.	.	PUNCT
bracis-19021	241	7	2944–2952	2944–2952	NUM
bracis-19021	241	8	(	(	PUNCT
bracis-19021	241	9	2015	2015	NUM
bracis-19021	241	10	)	)	PUNCT
bracis-19021	241	11	.	.	PUNCT
bracis-19021	242	1	http://papers.nips.cc/paper/5796-learning-continuous-control-policies-by-stochastic-value-gradients	http://papers.nips.cc/paper/5796-learning-continuous-control-policies-by-stochastic-value-gradients	PROPN
bracis-19021	242	2	henderson	henderson	PROPN
bracis-19021	242	3	,	,	PUNCT
bracis-19021	242	4	p.	p.	PROPN
bracis-19021	242	5	,	,	PUNCT
bracis-19021	242	6	islam	islam	PROPN
bracis-19021	242	7	,	,	PUNCT
bracis-19021	242	8	r.	r.	PROPN
bracis-19021	242	9	,	,	PUNCT
bracis-19021	242	10	bachman	bachman	PROPN
bracis-19021	242	11	,	,	PUNCT
bracis-19021	242	12	p.	p.	PROPN
bracis-19021	242	13	,	,	PUNCT
bracis-19021	242	14	pineau	pineau	PROPN
bracis-19021	242	15	,	,	PUNCT
bracis-19021	242	16	j.	j.	PROPN
bracis-19021	242	17	,	,	PUNCT
bracis-19021	242	18	precup	precup	PROPN
bracis-19021	242	19	,	,	PUNCT
bracis-19021	242	20	d.	d.	PROPN
bracis-19021	242	21	,	,	PUNCT
bracis-19021	242	22	meger	meger	PROPN
bracis-19021	242	23	,	,	PUNCT
bracis-19021	242	24	d.	d.	NOUN
bracis-19021	242	25	:	:	PUNCT
bracis-19021	242	26	deep	deep	ADJ
bracis-19021	242	27	reinforcement	reinforcement	NOUN
bracis-19021	242	28	learning	learning	NOUN
bracis-19021	242	29	that	that	PRON
bracis-19021	242	30	matters	matter	VERB
bracis-19021	242	31	.	.	PUNCT
bracis-19021	243	1	in	in	ADP
bracis-19021	243	2	:	:	PUNCT
bracis-19021	243	3	aaai	aaai	PROPN
bracis-19021	243	4	,	,	PUNCT
bracis-19021	243	5	pp	pp	ADJ
bracis-19021	243	6	.	.	PUNCT
bracis-19021	244	1	3207–3214	3207–3214	NUM
bracis-19021	244	2	.	.	PUNCT
bracis-19021	245	1	aaai	aaai	PROPN
bracis-19021	245	2	press	press	PROPN
bracis-19021	245	3	(	(	PUNCT
bracis-19021	245	4	2018	2018	NUM
bracis-19021	245	5	)	)	PUNCT
bracis-19021	245	6	google	google	PROPN
bracis-19021	245	7	scholar	scholar	NOUN
bracis-19021	245	8	  	  	SPACE
bracis-19021	245	9	ilyas	ilyas	PROPN
bracis-19021	245	10	,	,	PUNCT
bracis-19021	245	11	a.	a.	PROPN
bracis-19021	245	12	,	,	PUNCT
bracis-19021	245	13	et	et	PROPN
bracis-19021	245	14	al	al	PROPN
bracis-19021	245	15	.	.	PUNCT
bracis-19021	245	16	:	:	PUNCT
bracis-19021	246	1	a	a	DET
bracis-19021	246	2	closer	close	ADJ
bracis-19021	246	3	look	look	NOUN
bracis-19021	246	4	at	at	ADP
bracis-19021	246	5	deep	deep	ADJ
bracis-19021	246	6	policy	policy	NOUN
bracis-19021	246	7	gradients	gradient	NOUN
bracis-19021	246	8	.	.	PUNCT
bracis-19021	247	1	in	in	ADP
bracis-19021	247	2	:	:	PUNCT
bracis-19021	247	3	iclr	iclr	NOUN
bracis-19021	247	4	.	.	PUNCT
bracis-19021	248	1	openreview.net	openreview.net	X
bracis-19021	248	2	(	(	PUNCT
bracis-19021	248	3	2020	2020	NUM
bracis-19021	248	4	)	)	PUNCT
bracis-19021	248	5	google	google	PROPN
bracis-19021	248	6	scholar	scholar	NOUN
bracis-19021	248	7	  	  	SPACE
bracis-19021	248	8	islam	islam	PROPN
bracis-19021	248	9	,	,	PUNCT
bracis-19021	248	10	r.	r.	PROPN
bracis-19021	248	11	,	,	PUNCT
bracis-19021	248	12	henderson	henderson	PROPN
bracis-19021	248	13	,	,	PUNCT
bracis-19021	248	14	p.	p.	PROPN
bracis-19021	248	15	,	,	PUNCT
bracis-19021	248	16	gomrokchi	gomrokchi	NOUN
bracis-19021	248	17	,	,	PUNCT
bracis-19021	248	18	m.	m.	NOUN
bracis-19021	248	19	,	,	PUNCT
bracis-19021	248	20	precup	precup	PROPN
bracis-19021	248	21	,	,	PUNCT
bracis-19021	248	22	d.	d.	PROPN
bracis-19021	248	23	:	:	PUNCT
bracis-19021	248	24	reproducibility	reproducibility	NOUN
bracis-19021	248	25	of	of	ADP
bracis-19021	248	26	benchmarked	benchmarke	VERB
bracis-19021	248	27	deep	deep	ADJ
bracis-19021	248	28	reinforcement	reinforcement	NOUN
bracis-19021	248	29	learning	learning	NOUN
bracis-19021	248	30	tasks	task	NOUN
bracis-19021	248	31	for	for	ADP
bracis-19021	248	32	continuous	continuous	ADJ
bracis-19021	248	33	control	control	NOUN
bracis-19021	248	34	.	.	PUNCT
bracis-19021	249	1	corr	corr	PROPN
bracis-19021	249	2	arxiv:1708.04133	arxiv:1708.04133	PROPN
bracis-19021	249	3	(	(	PUNCT
bracis-19021	249	4	2017	2017	NUM
bracis-19021	249	5	)	)	PUNCT
bracis-19021	249	6	liu	liu	PROPN
bracis-19021	249	7	,	,	PUNCT
bracis-19021	249	8	z.	z.	PROPN
bracis-19021	249	9	,	,	PUNCT
bracis-19021	249	10	li	li	PROPN
bracis-19021	249	11	,	,	PUNCT
bracis-19021	249	12	x.	x.	PROPN
bracis-19021	249	13	,	,	PUNCT
bracis-19021	249	14	kang	kang	PROPN
bracis-19021	249	15	,	,	PUNCT
bracis-19021	249	16	b.	b.	PROPN
bracis-19021	249	17	,	,	PUNCT
bracis-19021	249	18	darrell	darrell	PROPN
bracis-19021	249	19	,	,	PUNCT
bracis-19021	249	20	t.	t.	PROPN
bracis-19021	249	21	:	:	PUNCT
bracis-19021	249	22	regularization	regularization	NOUN
bracis-19021	249	23	matters	matter	VERB
bracis-19021	249	24	for	for	ADP
bracis-19021	249	25	policy	policy	NOUN
bracis-19021	249	26	optimization	optimization	NOUN
bracis-19021	249	27	an	an	DET
bracis-19021	249	28	empirical	empirical	ADJ
bracis-19021	249	29	study	study	NOUN
bracis-19021	249	30	on	on	ADP
bracis-19021	249	31	continuous	continuous	ADJ
bracis-19021	249	32	control	control	NOUN
bracis-19021	249	33	.	.	PUNCT
bracis-19021	250	1	in	in	ADP
bracis-19021	250	2	:	:	PUNCT
bracis-19021	250	3	international	international	ADJ
bracis-19021	250	4	conference	conference	NOUN
bracis-19021	250	5	on	on	ADP
bracis-19021	250	6	learning	learn	VERB
bracis-19021	250	7	representations	representation	NOUN
bracis-19021	250	8	(	(	PUNCT
bracis-19021	250	9	2021	2021	NUM
bracis-19021	250	10	)	)	PUNCT
bracis-19021	250	11	.	.	PUNCT
bracis-19021	251	1	https://github.com/xuanlinli17/iclr2021_rlreg	https://github.com/xuanlinli17/iclr2021_rlreg	NOUN
bracis-19021	251	2	lovatto	lovatto	PROPN
bracis-19021	251	3	,	,	PUNCT
bracis-19021	251	4	a.g	a.g	PROPN
bracis-19021	251	5	.	.	PROPN
bracis-19021	251	6	,	,	PUNCT
bracis-19021	251	7	bueno	bueno	PROPN
bracis-19021	251	8	,	,	PUNCT
bracis-19021	251	9	t.p	t.p	PROPN
bracis-19021	251	10	.	.	PROPN
bracis-19021	251	11	,	,	PUNCT
bracis-19021	251	12	mauá	mauá	NOUN
bracis-19021	251	13	,	,	PUNCT
bracis-19021	251	14	d.d	d.d	PROPN
bracis-19021	251	15	.	.	PROPN
bracis-19021	251	16	,	,	PUNCT
bracis-19021	251	17	de	de	PROPN
bracis-19021	251	18	barros	barros	PROPN
bracis-19021	251	19	,	,	PUNCT
bracis-19021	251	20	l.n	l.n	PROPN
bracis-19021	251	21	.	.	PROPN
bracis-19021	251	22	:	:	PUNCT
bracis-19021	251	23	decision	decision	NOUN
bracis-19021	251	24	-	-	PUNCT
bracis-19021	251	25	aware	aware	ADJ
bracis-19021	251	26	model	model	NOUN
bracis-19021	251	27	learning	learn	VERB
bracis-19021	251	28	for	for	ADP
bracis-19021	251	29	actor	actor	NOUN
bracis-19021	251	30	-	-	PUNCT
bracis-19021	251	31	critic	critic	NOUN
bracis-19021	251	32	methods	method	NOUN
bracis-19021	251	33	:	:	PUNCT
bracis-19021	251	34	when	when	SCONJ
bracis-19021	251	35	theory	theory	NOUN
bracis-19021	251	36	does	do	AUX
bracis-19021	251	37	not	not	PART
bracis-19021	251	38	meet	meet	VERB
bracis-19021	251	39	practice	practice	NOUN
bracis-19021	251	40	.	.	PUNCT
bracis-19021	252	1	in	in	ADP
bracis-19021	252	2	:	:	PUNCT
bracis-19021	252	3	proceedings	proceeding	NOUN
bracis-19021	252	4	on	on	ADP
bracis-19021	252	5	“	"	PUNCT
bracis-19021	252	6	i	i	PRON
bracis-19021	252	7	ca	can	AUX
bracis-19021	252	8	n’t	not	PART
bracis-19021	252	9	believe	believe	VERB
bracis-19021	252	10	it	it	PRON
bracis-19021	252	11	’s	’	VERB
bracis-19021	252	12	not	not	PART
bracis-19021	252	13	better	well	ADJ
bracis-19021	252	14	!	!	PUNCT
bracis-19021	252	15	”	"	PUNCT
bracis-19021	253	1	at	at	ADP
bracis-19021	253	2	neurips	neurip	NOUN
bracis-19021	253	3	workshops	workshop	NOUN
bracis-19021	253	4	.	.	PUNCT
bracis-19021	254	1	proceedings	proceeding	NOUN
bracis-19021	254	2	of	of	ADP
bracis-19021	254	3	machine	machine	NOUN
bracis-19021	254	4	learning	learn	VERB
bracis-19021	254	5	research	research	NOUN
bracis-19021	254	6	,	,	PUNCT
bracis-19021	254	7	vol	vol	NOUN
bracis-19021	254	8	.	.	PROPN
bracis-19021	254	9	137	137	NUM
bracis-19021	254	10	,	,	PUNCT
bracis-19021	254	11	pp	pp	ADJ
bracis-19021	254	12	.	.	PUNCT
bracis-19021	255	1	76–86	76–86	NUM
bracis-19021	255	2	.	.	PUNCT
bracis-19021	255	3	pmlr	pmlr	NOUN
bracis-19021	255	4	,	,	PUNCT
bracis-19021	255	5	december	december	PROPN
bracis-19021	255	6	2020	2020	NUM
bracis-19021	255	7	.	.	PUNCT
bracis-19021	256	1	http://proceedings.mlr.press/v137/lovatto20a.html	http://proceedings.mlr.press/v137/lovatto20a.html	PROPN
bracis-19021	256	2	mnih	mnih	PROPN
bracis-19021	256	3	,	,	PUNCT
bracis-19021	256	4	v.	v.	PROPN
bracis-19021	256	5	,	,	PUNCT
bracis-19021	256	6	et	et	PROPN
bracis-19021	256	7	al	al	PROPN
bracis-19021	256	8	.	.	PROPN
bracis-19021	256	9	:	:	PUNCT
bracis-19021	256	10	human	human	ADJ
bracis-19021	256	11	-	-	PUNCT
bracis-19021	256	12	level	level	NOUN
bracis-19021	256	13	control	control	NOUN
bracis-19021	256	14	through	through	ADP
bracis-19021	256	15	deep	deep	ADJ
bracis-19021	256	16	reinforcement	reinforcement	NOUN
bracis-19021	256	17	learning	learning	NOUN
bracis-19021	256	18	.	.	PUNCT
bracis-19021	257	1	nature	nature	NOUN
bracis-19021	257	2	518(7540	518(7540	NUM
bracis-19021	257	3	)	)	PUNCT
bracis-19021	257	4	,	,	PUNCT
bracis-19021	257	5	529–533	529–533	NUM
bracis-19021	257	6	(	(	PUNCT
bracis-19021	257	7	2015	2015	NUM
bracis-19021	257	8	)	)	PUNCT
bracis-19021	257	9	article	article	NOUN
bracis-19021	257	10	  	  	SPACE
bracis-19021	257	11	google	google	PROPN
bracis-19021	257	12	scholar	scholar	NOUN
bracis-19021	257	13	  	  	SPACE
bracis-19021	257	14	moerland	moerland	NOUN
bracis-19021	257	15	,	,	PUNCT
bracis-19021	257	16	t.m	t.m	PROPN
bracis-19021	257	17	.	.	PROPN
bracis-19021	257	18	,	,	PUNCT
bracis-19021	257	19	broekens	broekens	PROPN
bracis-19021	257	20	,	,	PUNCT
bracis-19021	257	21	j.	j.	PROPN
bracis-19021	257	22	,	,	PUNCT
bracis-19021	257	23	jonker	jonker	PROPN
bracis-19021	257	24	,	,	PUNCT
bracis-19021	257	25	c.m	c.m	PROPN
bracis-19021	257	26	.	.	PROPN
bracis-19021	257	27	:	:	PUNCT
bracis-19021	257	28	model	model	NOUN
bracis-19021	257	29	-	-	PUNCT
bracis-19021	257	30	based	base	VERB
bracis-19021	257	31	reinforcement	reinforcement	NOUN
bracis-19021	257	32	learning	learning	NOUN
bracis-19021	257	33	:	:	PUNCT
bracis-19021	257	34	a	a	DET
bracis-19021	257	35	survey	survey	NOUN
bracis-19021	257	36	.	.	PUNCT
bracis-19021	258	1	in	in	ADP
bracis-19021	258	2	:	:	PUNCT
bracis-19021	258	3	proceedings	proceeding	NOUN
bracis-19021	258	4	of	of	ADP
bracis-19021	258	5	the	the	DET
bracis-19021	258	6	international	international	ADJ
bracis-19021	258	7	conference	conference	NOUN
bracis-19021	258	8	on	on	ADP
bracis-19021	258	9	electronic	electronic	ADJ
bracis-19021	258	10	business	business	NOUN
bracis-19021	258	11	(	(	PUNCT
bracis-19021	258	12	iceb	iceb	NOUN
bracis-19021	258	13	)	)	PUNCT
bracis-19021	258	14	2018	2018	NUM
bracis-19021	258	15	-	-	PUNCT
bracis-19021	258	16	december	december	PROPN
bracis-19021	258	17	,	,	PUNCT
bracis-19021	258	18	pp	pp	ADJ
bracis-19021	258	19	.	.	PUNCT
bracis-19021	259	1	421–429	421–429	NUM
bracis-19021	259	2	(	(	PUNCT
bracis-19021	259	3	2020	2020	NUM
bracis-19021	259	4	)	)	PUNCT
bracis-19021	259	5	.	.	PUNCT
bracis-19021	260	1	http://arxiv.org/abs/2006.16712	http://arxiv.org/abs/2006.16712	PROPN
bracis-19021	260	2	mohamed	mohamed	PROPN
bracis-19021	260	3	,	,	PUNCT
bracis-19021	260	4	s.	s.	PROPN
bracis-19021	260	5	,	,	PUNCT
bracis-19021	260	6	rosca	rosca	PROPN
bracis-19021	260	7	,	,	PUNCT
bracis-19021	260	8	m.	m.	NOUN
bracis-19021	260	9	,	,	PUNCT
bracis-19021	260	10	figurnov	figurnov	PROPN
bracis-19021	260	11	,	,	PUNCT
bracis-19021	260	12	m.	m.	NOUN
bracis-19021	260	13	,	,	PUNCT
bracis-19021	260	14	mnih	mnih	PROPN
bracis-19021	260	15	,	,	PUNCT
bracis-19021	260	16	a.	a.	NOUN
bracis-19021	260	17	:	:	PUNCT
bracis-19021	260	18	monte	monte	PROPN
bracis-19021	260	19	carlo	carlo	PROPN
bracis-19021	260	20	gradient	gradient	PROPN
bracis-19021	260	21	estimation	estimation	NOUN
bracis-19021	260	22	in	in	ADP
bracis-19021	260	23	machine	machine	NOUN
bracis-19021	260	24	learning	learning	NOUN
bracis-19021	260	25	.	.	PUNCT
bracis-19021	261	1	j.	j.	PROPN
bracis-19021	261	2	mach	mach	PROPN
bracis-19021	261	3	.	.	PUNCT
bracis-19021	262	1	learn	learn	VERB
bracis-19021	262	2	.	.	PUNCT
bracis-19021	263	1	res	re	NOUN
bracis-19021	263	2	.	.	PROPN
bracis-19021	264	1	21	21	NUM
bracis-19021	264	2	,	,	PUNCT
bracis-19021	264	3	132:1–132:62	132:1–132:62	NUM
bracis-19021	264	4	(	(	PUNCT
bracis-19021	264	5	2020	2020	NUM
bracis-19021	264	6	)	)	PUNCT
bracis-19021	264	7	google	google	PROPN
bracis-19021	264	8	scholar	scholar	NOUN
bracis-19021	264	9	  	  	SPACE
bracis-19021	264	10	paszke	paszke	NOUN
bracis-19021	264	11	,	,	PUNCT
bracis-19021	264	12	a.	a.	NOUN
bracis-19021	264	13	,	,	PUNCT
bracis-19021	264	14	et	et	PROPN
bracis-19021	264	15	al	al	PROPN
bracis-19021	264	16	.	.	PUNCT
bracis-19021	264	17	:	:	PUNCT
bracis-19021	265	1	pytorch	pytorch	NOUN
bracis-19021	265	2	:	:	PUNCT
bracis-19021	265	3	an	an	DET
bracis-19021	265	4	imperative	imperative	ADJ
bracis-19021	265	5	style	style	NOUN
bracis-19021	265	6	,	,	PUNCT
bracis-19021	265	7	high	high	ADJ
bracis-19021	265	8	-	-	PUNCT
bracis-19021	265	9	performance	performance	NOUN
bracis-19021	265	10	deep	deep	ADJ
bracis-19021	265	11	learning	learning	NOUN
bracis-19021	265	12	library	library	NOUN
bracis-19021	265	13	.	.	PUNCT
bracis-19021	266	1	in	in	ADP
bracis-19021	266	2	:	:	PUNCT
bracis-19021	266	3	advances	advance	NOUN
bracis-19021	266	4	in	in	ADP
bracis-19021	266	5	neural	neural	ADJ
bracis-19021	266	6	information	information	NOUN
bracis-19021	266	7	processing	processing	NOUN
bracis-19021	266	8	systems	system	NOUN
bracis-19021	266	9	,	,	PUNCT
bracis-19021	266	10	vol	vol	NOUN
bracis-19021	266	11	.	.	PROPN
bracis-19021	267	1	32	32	NUM
bracis-19021	267	2	,	,	PUNCT
bracis-19021	267	3	pp	pp	ADJ
bracis-19021	267	4	.	.	PUNCT
bracis-19021	268	1	8024–8035	8024–8035	NUM
bracis-19021	268	2	.	.	PUNCT
bracis-19021	269	1	curran	curran	PROPN
bracis-19021	269	2	associates	associates	PROPN
bracis-19021	269	3	,	,	PUNCT
bracis-19021	269	4	inc	inc	PROPN
bracis-19021	269	5	.	.	PROPN
bracis-19021	269	6	(	(	PUNCT
bracis-19021	269	7	2019	2019	NUM
bracis-19021	269	8	)	)	PUNCT
bracis-19021	269	9	.	.	PUNCT
bracis-19021	270	1	http://papers.nips.cc/paper/9015-pytorch-an-imperative-style-high-performance-deep-learning-library.pdf	http://papers.nips.cc/paper/9015-pytorch-an-imperative-style-high-performance-deep-learning-library.pdf	PROPN
bracis-19021	270	2	pedregosa	pedregosa	PROPN
bracis-19021	270	3	,	,	PUNCT
bracis-19021	270	4	f.	f.	PROPN
bracis-19021	270	5	,	,	PUNCT
bracis-19021	270	6	et	et	PROPN
bracis-19021	270	7	al	al	PROPN
bracis-19021	270	8	.	.	PROPN
bracis-19021	270	9	:	:	PUNCT
bracis-19021	271	1	scikit	scikit	NOUN
bracis-19021	271	2	-	-	PUNCT
bracis-19021	271	3	learn	learn	VERB
bracis-19021	271	4	:	:	PUNCT
bracis-19021	271	5	machine	machine	NOUN
bracis-19021	271	6	learning	learning	NOUN
bracis-19021	271	7	in	in	ADP
bracis-19021	271	8	python	python	PROPN
bracis-19021	271	9	.	.	PUNCT
bracis-19021	272	1	j.	j.	PROPN
bracis-19021	272	2	mach	mach	PROPN
bracis-19021	272	3	.	.	PUNCT
bracis-19021	273	1	learn	learn	VERB
bracis-19021	273	2	.	.	PUNCT
bracis-19021	274	1	res	re	NOUN
bracis-19021	274	2	.	.	PROPN
bracis-19021	275	1	12	12	NUM
bracis-19021	275	2	,	,	PUNCT
bracis-19021	275	3	2825–2830	2825–2830	NUM
bracis-19021	275	4	(	(	PUNCT
bracis-19021	275	5	2011	2011	NUM
bracis-19021	275	6	)	)	PUNCT
bracis-19021	275	7	mathscinet	mathscinet	NOUN
bracis-19021	275	8	  	  	SPACE
bracis-19021	275	9	math	math	NOUN
bracis-19021	275	10	  	  	SPACE
bracis-19021	275	11	google	google	PROPN
bracis-19021	275	12	scholar	scholar	NOUN
bracis-19021	275	13	  	  	SPACE
bracis-19021	275	14	polydoros	polydoro	NOUN
bracis-19021	275	15	,	,	PUNCT
bracis-19021	275	16	a.s	a.s	PROPN
bracis-19021	275	17	.	.	PROPN
bracis-19021	275	18	,	,	PUNCT
bracis-19021	275	19	nalpantidis	nalpantidis	PROPN
bracis-19021	275	20	,	,	PUNCT
bracis-19021	275	21	l.	l.	PROPN
bracis-19021	275	22	:	:	PUNCT
bracis-19021	275	23	survey	survey	NOUN
bracis-19021	275	24	of	of	ADP
bracis-19021	275	25	model	model	NOUN
bracis-19021	275	26	-	-	PUNCT
bracis-19021	275	27	based	base	VERB
bracis-19021	275	28	reinforcement	reinforcement	NOUN
bracis-19021	275	29	learning	learning	NOUN
bracis-19021	275	30	:	:	PUNCT
bracis-19021	275	31	applications	application	NOUN
bracis-19021	275	32	on	on	ADP
bracis-19021	275	33	robotics	robotic	NOUN
bracis-19021	275	34	.	.	PUNCT
bracis-19021	276	1	j.	j.	PROPN
bracis-19021	276	2	intell	intell	PROPN
bracis-19021	276	3	.	.	PUNCT
bracis-19021	277	1	robot	robot	NOUN
bracis-19021	277	2	.	.	PUNCT
bracis-19021	278	1	syst	syst	PROPN
bracis-19021	278	2	.	.	PUNCT
bracis-19021	279	1	86(2	86(2	PROPN
bracis-19021	279	2	)	)	PUNCT
bracis-19021	279	3	,	,	PUNCT
bracis-19021	279	4	153–173	153–173	NUM
bracis-19021	279	5	(	(	PUNCT
bracis-19021	279	6	2017	2017	NUM
bracis-19021	279	7	)	)	PUNCT
bracis-19021	279	8	.	.	PUNCT
bracis-19021	280	1	https://doi.org/10.1007/s10846-017-0468-y	https://doi.org/10.1007/s10846-017-0468-y	PROPN
bracis-19021	280	2	article	article	NOUN
bracis-19021	280	3	  	  	SPACE
bracis-19021	280	4	google	google	PROPN
bracis-19021	280	5	scholar	scholar	NOUN
bracis-19021	280	6	  	  	SPACE
bracis-19021	280	7	recht	recht	NOUN
bracis-19021	280	8	,	,	PUNCT
bracis-19021	280	9	b.	b.	PROPN
bracis-19021	280	10	:	:	PUNCT
bracis-19021	280	11	a	a	DET
bracis-19021	280	12	tour	tour	NOUN
bracis-19021	280	13	of	of	ADP
bracis-19021	280	14	reinforcement	reinforcement	NOUN
bracis-19021	280	15	learning	learning	NOUN
bracis-19021	280	16	:	:	PUNCT
bracis-19021	280	17	the	the	DET
bracis-19021	280	18	view	view	NOUN
bracis-19021	280	19	from	from	ADP
bracis-19021	280	20	continuous	continuous	ADJ
bracis-19021	280	21	control	control	NOUN
bracis-19021	280	22	.	.	PUNCT
bracis-19021	281	1	ann	ann	AUX
bracis-19021	281	2	.	.	PUNCT
bracis-19021	281	3	rev	rev	PROPN
bracis-19021	281	4	.	.	PROPN
bracis-19021	281	5	control	control	PROPN
bracis-19021	281	6	robot	robot	PROPN
bracis-19021	281	7	.	.	PUNCT
bracis-19021	282	1	auton	auton	PROPN
bracis-19021	282	2	.	.	PUNCT
bracis-19021	283	1	syst	syst	PROPN
bracis-19021	283	2	.	.	PUNCT
bracis-19021	284	1	2(1	2(1	NUM
bracis-19021	284	2	)	)	PUNCT
bracis-19021	284	3	,	,	PUNCT
bracis-19021	284	4	253–279	253–279	NUM
bracis-19021	284	5	(	(	PUNCT
bracis-19021	284	6	2019	2019	NUM
bracis-19021	284	7	)	)	PUNCT
bracis-19021	284	8	.	.	PUNCT
bracis-19021	285	1	https://doi.org/10.1146/annurev-control-053018-023825	https://doi.org/10.1146/annurev-control-053018-023825	PROPN
bracis-19021	285	2	,	,	PUNCT
bracis-19021	285	3	http://arxiv.org/abs/1806.09460	http://arxiv.org/abs/1806.09460	NOUN
bracis-19021	285	4	ruder	ruder	PROPN
bracis-19021	285	5	,	,	PUNCT
bracis-19021	285	6	s.	s.	PROPN
bracis-19021	285	7	:	:	PUNCT
bracis-19021	285	8	an	an	DET
bracis-19021	285	9	overview	overview	NOUN
bracis-19021	285	10	of	of	ADP
bracis-19021	285	11	gradient	gradient	ADJ
bracis-19021	285	12	descent	descent	NOUN
bracis-19021	285	13	optimization	optimization	NOUN
bracis-19021	285	14	algorithms	algorithm	NOUN
bracis-19021	285	15	.	.	PUNCT
bracis-19021	286	1	corr	corr	PROPN
bracis-19021	286	2	arxiv:1609.04747	arxiv:1609.04747	PROPN
bracis-19021	286	3	(	(	PUNCT
bracis-19021	286	4	2016	2016	NUM
bracis-19021	286	5	)	)	PUNCT
bracis-19021	286	6	schulman	schulman	PROPN
bracis-19021	286	7	,	,	PUNCT
bracis-19021	286	8	j.	j.	PROPN
bracis-19021	286	9	,	,	PUNCT
bracis-19021	286	10	heess	heess	PROPN
bracis-19021	286	11	,	,	PUNCT
bracis-19021	286	12	n.	n.	PROPN
bracis-19021	286	13	,	,	PUNCT
bracis-19021	286	14	weber	weber	PROPN
bracis-19021	286	15	,	,	PUNCT
bracis-19021	286	16	t.	t.	PROPN
bracis-19021	286	17	,	,	PUNCT
bracis-19021	286	18	abbeel	abbeel	NOUN
bracis-19021	286	19	,	,	PUNCT
bracis-19021	286	20	p.	p.	NOUN
bracis-19021	286	21	:	:	PUNCT
bracis-19021	286	22	gradient	gradient	ADJ
bracis-19021	286	23	estimation	estimation	NOUN
bracis-19021	286	24	using	use	VERB
bracis-19021	286	25	stochastic	stochastic	ADJ
bracis-19021	286	26	computation	computation	NOUN
bracis-19021	286	27	graphs	graph	NOUN
bracis-19021	286	28	.	.	PUNCT
bracis-19021	287	1	in	in	ADP
bracis-19021	287	2	:	:	PUNCT
bracis-19021	287	3	nips	nip	NOUN
bracis-19021	287	4	,	,	PUNCT
bracis-19021	287	5	pp	pp	ADV
bracis-19021	287	6	.	.	PUNCT
bracis-19021	288	1	3528–3536	3528–3536	NUM
bracis-19021	288	2	(	(	PUNCT
bracis-19021	288	3	2015	2015	NUM
bracis-19021	288	4	)	)	PUNCT
bracis-19021	288	5	google	google	NOUN
bracis-19021	288	6	scholar	scholar	NOUN
bracis-19021	288	7	  	  	SPACE
bracis-19021	288	8	silver	silver	NOUN
bracis-19021	288	9	,	,	PUNCT
bracis-19021	288	10	d.	d.	PROPN
bracis-19021	288	11	,	,	PUNCT
bracis-19021	288	12	lever	lever	PROPN
bracis-19021	288	13	,	,	PUNCT
bracis-19021	288	14	g.	g.	PROPN
bracis-19021	288	15	,	,	PUNCT
bracis-19021	288	16	technologies	technology	NOUN
bracis-19021	288	17	,	,	PUNCT
bracis-19021	288	18	d.	d.	PROPN
bracis-19021	288	19	,	,	PUNCT
bracis-19021	288	20	lever	lever	ADJ
bracis-19021	288	21	,	,	PUNCT
bracis-19021	288	22	g.u.y	g.u.y	NOUN
bracis-19021	288	23	.	.	PROPN
bracis-19021	288	24	,	,	PUNCT
bracis-19021	288	25	ac	ac	PROPN
bracis-19021	288	26	,	,	PUNCT
bracis-19021	288	27	u.c.l	u.c.l	NOUN
bracis-19021	288	28	.	.	PUNCT
bracis-19021	288	29	:	:	PUNCT
bracis-19021	289	1	deterministic	deterministic	ADJ
bracis-19021	289	2	policy	policy	NOUN
bracis-19021	289	3	gradient	gradient	NOUN
bracis-19021	289	4	(	(	PUNCT
bracis-19021	289	5	dpg	dpg	PROPN
bracis-19021	289	6	)	)	PUNCT
bracis-19021	289	7	.	.	PUNCT
bracis-19021	290	1	in	in	ADP
bracis-19021	290	2	:	:	PUNCT
bracis-19021	290	3	proceedings	proceeding	NOUN
bracis-19021	290	4	of	of	ADP
bracis-19021	290	5	the	the	DET
bracis-19021	290	6	31st	31st	ADJ
bracis-19021	290	7	international	international	ADJ
bracis-19021	290	8	conference	conference	NOUN
bracis-19021	290	9	on	on	ADP
bracis-19021	290	10	machine	machine	NOUN
bracis-19021	290	11	learning	learning	NOUN
bracis-19021	290	12	,	,	PUNCT
bracis-19021	290	13	vol	vol	NOUN
bracis-19021	290	14	.	.	PROPN
bracis-19021	290	15	32	32	NUM
bracis-19021	290	16	,	,	PUNCT
bracis-19021	290	17	no	no	INTJ
bracis-19021	290	18	.	.	NOUN
bracis-19021	290	19	1	1	NUM
bracis-19021	290	20	,	,	PUNCT
bracis-19021	290	21	pp	pp	ADJ
bracis-19021	290	22	.	.	PUNCT
bracis-19021	291	1	387–395	387–395	NUM
bracis-19021	291	2	(	(	PUNCT
bracis-19021	291	3	2014	2014	NUM
bracis-19021	291	4	)	)	PUNCT
bracis-19021	291	5	.	.	PUNCT
bracis-19021	292	1	http://proceedings.mlr.press/v32/silver14.html	http://proceedings.mlr.press/v32/silver14.html	PROPN
bracis-19021	292	2	szepesvári	szepesvári	NOUN
bracis-19021	292	3	,	,	PUNCT
bracis-19021	292	4	c.	c.	NOUN
bracis-19021	292	5	:	:	PUNCT
bracis-19021	292	6	algorithms	algorithm	NOUN
bracis-19021	292	7	for	for	ADP
bracis-19021	292	8	reinforcement	reinforcement	NOUN
bracis-19021	292	9	learning	learning	NOUN
bracis-19021	292	10	.	.	PUNCT
bracis-19021	293	1	synthesis	synthesis	NOUN
bracis-19021	293	2	lectures	lecture	NOUN
bracis-19021	293	3	on	on	ADP
bracis-19021	293	4	artificial	artificial	ADJ
bracis-19021	293	5	intelligence	intelligence	NOUN
bracis-19021	293	6	and	and	CCONJ
bracis-19021	293	7	machine	machine	NOUN
bracis-19021	293	8	learning	learning	NOUN
bracis-19021	293	9	.	.	PUNCT
bracis-19021	294	1	morgan	morgan	PROPN
bracis-19021	294	2	&	&	CCONJ
bracis-19021	294	3	claypool	claypool	PROPN
bracis-19021	294	4	publishers	publisher	NOUN
bracis-19021	294	5	(	(	PUNCT
bracis-19021	294	6	2010	2010	NUM
bracis-19021	294	7	)	)	PUNCT
bracis-19021	294	8	.	.	PUNCT
bracis-19021	295	1	https://doi.org/10.2200/s00268ed1v01y201005aim009	https://doi.org/10.2200/s00268ed1v01y201005aim009	NOUN
bracis-19021	295	2	tits	tit	NOUN
bracis-19021	295	3	,	,	PUNCT
bracis-19021	295	4	a.l	a.l	PROPN
bracis-19021	295	5	.	.	PROPN
bracis-19021	295	6	,	,	PUNCT
bracis-19021	295	7	yang	yang	PROPN
bracis-19021	295	8	,	,	PUNCT
bracis-19021	295	9	y.	y.	PROPN
bracis-19021	295	10	:	:	PUNCT
bracis-19021	295	11	globally	globally	ADV
bracis-19021	295	12	convergent	convergent	ADJ
bracis-19021	295	13	algorithms	algorithm	NOUN
bracis-19021	295	14	for	for	ADP
bracis-19021	295	15	robust	robust	ADJ
bracis-19021	295	16	pole	pole	NOUN
bracis-19021	295	17	assignment	assignment	NOUN
bracis-19021	295	18	by	by	ADP
bracis-19021	295	19	state	state	NOUN
bracis-19021	295	20	feedback	feedback	NOUN
bracis-19021	295	21	.	.	PUNCT
bracis-19021	296	1	ieee	ieee	PROPN
bracis-19021	296	2	trans	trans	PROPN
bracis-19021	296	3	.	.	PUNCT
bracis-19021	297	1	autom	autom	PROPN
bracis-19021	297	2	.	.	PUNCT
bracis-19021	298	1	control	control	PROPN
bracis-19021	298	2	41(10	41(10	PROPN
bracis-19021	298	3	)	)	PUNCT
bracis-19021	298	4	,	,	PUNCT
bracis-19021	298	5	1432–1452	1432–1452	NUM
bracis-19021	298	6	(	(	PUNCT
bracis-19021	298	7	1996	1996	NUM
bracis-19021	298	8	)	)	PUNCT
bracis-19021	298	9	.	.	PUNCT
bracis-19021	299	1	https://doi.org/10.1109/9.539425	https://doi.org/10.1109/9.539425	PROPN
bracis-19021	299	2	article	article	NOUN
bracis-19021	299	3	  	  	SPACE
bracis-19021	299	4	mathscinet	mathscinet	NOUN
bracis-19021	299	5	  	  	SPACE
bracis-19021	299	6	math	math	NOUN
bracis-19021	299	7	  	  	SPACE
bracis-19021	299	8	google	google	PROPN
bracis-19021	299	9	scholar	scholar	NOUN
bracis-19021	299	10	  	  	SPACE
bracis-19021	299	11	todorov	todorov	NOUN
bracis-19021	299	12	,	,	PUNCT
bracis-19021	299	13	e.	e.	PROPN
bracis-19021	299	14	:	:	PUNCT
bracis-19021	299	15	optimal	optimal	ADJ
bracis-19021	299	16	control	control	NOUN
bracis-19021	299	17	theory	theory	NOUN
bracis-19021	299	18	.	.	PUNCT
bracis-19021	300	1	bayesian	bayesian	NOUN
bracis-19021	300	2	brain	brain	NOUN
bracis-19021	300	3	:	:	PUNCT
bracis-19021	300	4	probabilistic	probabilistic	ADJ
bracis-19021	300	5	approaches	approach	NOUN
bracis-19021	300	6	to	to	ADP
bracis-19021	300	7	neural	neural	ADJ
bracis-19021	300	8	coding	coding	NOUN
bracis-19021	300	9	,	,	PUNCT
bracis-19021	300	10	pp	pp	ADJ
bracis-19021	300	11	.	.	PUNCT
bracis-19021	301	1	269–298	269–298	NUM
bracis-19021	301	2	(	(	PUNCT
bracis-19021	301	3	2006	2006	NUM
bracis-19021	301	4	)	)	PUNCT
bracis-19021	301	5	google	google	PROPN
bracis-19021	301	6	scholar	scholar	NOUN
bracis-19021	301	7	  	  	SPACE
bracis-19021	301	8	todorov	todorov	NOUN
bracis-19021	301	9	,	,	PUNCT
bracis-19021	301	10	e.	e.	PROPN
bracis-19021	301	11	,	,	PUNCT
bracis-19021	301	12	erez	erez	PROPN
bracis-19021	301	13	,	,	PUNCT
bracis-19021	301	14	t.	t.	PROPN
bracis-19021	301	15	,	,	PUNCT
bracis-19021	301	16	tassa	tassa	PROPN
bracis-19021	301	17	,	,	PUNCT
bracis-19021	301	18	y.	y.	NOUN
bracis-19021	301	19	:	:	PUNCT
bracis-19021	301	20	mujoco	mujoco	ADJ
bracis-19021	301	21	:	:	PUNCT
bracis-19021	301	22	a	a	DET
bracis-19021	301	23	physics	physics	NOUN
bracis-19021	301	24	engine	engine	NOUN
bracis-19021	301	25	for	for	ADP
bracis-19021	301	26	model	model	NOUN
bracis-19021	301	27	-	-	PUNCT
bracis-19021	301	28	based	base	VERB
bracis-19021	301	29	control	control	NOUN
bracis-19021	301	30	.	.	PUNCT
bracis-19021	302	1	in	in	ADP
bracis-19021	302	2	:	:	PUNCT
bracis-19021	302	3	ieee	ieee	PROPN
bracis-19021	302	4	international	international	ADJ
bracis-19021	302	5	conference	conference	NOUN
bracis-19021	302	6	on	on	ADP
bracis-19021	302	7	intelligent	intelligent	ADJ
bracis-19021	302	8	robots	robot	NOUN
bracis-19021	302	9	and	and	CCONJ
bracis-19021	302	10	systems	system	NOUN
bracis-19021	302	11	,	,	PUNCT
bracis-19021	302	12	pp	pp	ADJ
bracis-19021	302	13	.	.	PUNCT
bracis-19021	303	1	5026–5033	5026–5033	NUM
bracis-19021	303	2	.	.	X
bracis-19021	303	3	ieee	ieee	NOUN
bracis-19021	303	4	(	(	PUNCT
bracis-19021	303	5	2012	2012	NUM
bracis-19021	303	6	)	)	PUNCT
bracis-19021	303	7	.	.	PUNCT
bracis-19021	304	1	https://doi.org/10.1109/iros.2012.6386109	https://doi.org/10.1109/iros.2012.6386109	PROPN
bracis-19021	304	2	,	,	PUNCT
bracis-19021	304	3	http://ieeexplore.ieee.org/document/6386109/	http://ieeexplore.ieee.org/document/6386109/	ADJ
bracis-19021	304	4	virtanen	virtanen	NOUN
bracis-19021	304	5	,	,	PUNCT
bracis-19021	304	6	p.	p.	PROPN
bracis-19021	304	7	,	,	PUNCT
bracis-19021	304	8	et	et	PROPN
bracis-19021	304	9	al	al	PROPN
bracis-19021	304	10	.	.	PUNCT
bracis-19021	304	11	:	:	PUNCT
bracis-19021	305	1	scipy	scipy	PROPN
bracis-19021	305	2	1.0	1.0	NUM
bracis-19021	305	3	:	:	PUNCT
bracis-19021	305	4	fundamental	fundamental	ADJ
bracis-19021	305	5	algorithms	algorithm	NOUN
bracis-19021	305	6	for	for	ADP
bracis-19021	305	7	scientific	scientific	ADJ
bracis-19021	305	8	computing	computing	NOUN
bracis-19021	305	9	in	in	ADP
bracis-19021	305	10	python	python	PROPN
bracis-19021	305	11	.	.	PUNCT
bracis-19021	306	1	nat	nat	PROPN
bracis-19021	306	2	.	.	PUNCT
bracis-19021	307	1	meth	meth	PROPN
bracis-19021	307	2	.	.	PUNCT
bracis-19021	308	1	17	17	NUM
bracis-19021	308	2	,	,	PUNCT
bracis-19021	308	3	261–272	261–272	NUM
bracis-19021	308	4	(	(	PUNCT
bracis-19021	308	5	2020	2020	NUM
bracis-19021	308	6	)	)	PUNCT
bracis-19021	308	7	.	.	PUNCT
bracis-19021	309	1	https://doi.org/10.1038/s41592-019-0686-2	https://doi.org/10.1038/s41592-019-0686-2	PROPN
bracis-19021	309	2	article	article	PROPN
bracis-19021	309	3	  	  	SPACE
bracis-19021	309	4	google	google	PROPN
bracis-19021	309	5	scholar	scholar	NOUN
bracis-19021	309	6	  	  	SPACE
bracis-19021	309	7	download	download	NOUN
bracis-19021	309	8	references	reference	NOUN
bracis-19021	309	9	acknowledgments	acknowledgment	NOUN
bracis-19021	309	10	this	this	DET
bracis-19021	309	11	work	work	NOUN
bracis-19021	309	12	was	be	AUX
bracis-19021	309	13	partly	partly	ADV
bracis-19021	309	14	supported	support	VERB
bracis-19021	309	15	by	by	ADP
bracis-19021	309	16	the	the	DET
bracis-19021	309	17	capes	cape	NOUN
bracis-19021	309	18	grant	grant	VERB
bracis-19021	309	19	88887.339578/2019	88887.339578/2019	NUM
bracis-19021	309	20	-	-	SYM
bracis-19021	309	21	00	00	PUNCT
bracis-19021	310	1	(	(	PUNCT
bracis-19021	310	2	first	first	ADJ
bracis-19021	310	3	author	author	NOUN
bracis-19021	310	4	)	)	PUNCT
bracis-19021	310	5	,	,	PUNCT
bracis-19021	310	6	fapesp	fapesp	NOUN
bracis-19021	310	7	grant	grant	NOUN
bracis-19021	310	8	2016/22900	2016/22900	NUM
bracis-19021	310	9	-	-	SYM
bracis-19021	310	10	1	1	NUM
bracis-19021	310	11	(	(	PUNCT
bracis-19021	310	12	second	second	ADJ
bracis-19021	310	13	author	author	NOUN
bracis-19021	310	14	)	)	PUNCT
bracis-19021	310	15	,	,	PUNCT
bracis-19021	310	16	and	and	CCONJ
bracis-19021	310	17	cnpq	cnpq	NOUN
bracis-19021	310	18	scholarship	scholarship	NOUN
bracis-19021	310	19	307979/2018	307979/2018	NUM
bracis-19021	310	20	-	-	SYM
bracis-19021	310	21	0	0	NUM
bracis-19021	310	22	(	(	PUNCT
bracis-19021	310	23	third	third	ADJ
bracis-19021	310	24	author	author	NOUN
bracis-19021	310	25	)	)	PUNCT
bracis-19021	310	26	.	.	PUNCT
bracis-19021	311	1	author	author	NOUN
bracis-19021	311	2	information	information	NOUN
bracis-19021	311	3	authors	author	NOUN
bracis-19021	311	4	and	and	CCONJ
bracis-19021	311	5	affiliations	affiliation	NOUN
bracis-19021	311	6	instituto	instituto	PROPN
bracis-19021	311	7	de	de	PROPN
bracis-19021	311	8	matemática	matemática	PROPN
bracis-19021	311	9	e	e	PROPN
bracis-19021	311	10	estatística	estatística	PROPN
bracis-19021	311	11	,	,	PUNCT
bracis-19021	311	12	universidade	universidade	PROPN
bracis-19021	311	13	de	de	PROPN
bracis-19021	311	14	são	são	PROPN
bracis-19021	311	15	paulo	paulo	PROPN
bracis-19021	311	16	,	,	PUNCT
bracis-19021	311	17	são	são	PROPN
bracis-19021	311	18	paulo	paulo	PROPN
bracis-19021	311	19	,	,	PUNCT
bracis-19021	311	20	brazil	brazil	PROPN
bracis-19021	311	21	ângelo	ângelo	VERB
bracis-19021	311	22	gregório	gregório	PROPN
bracis-19021	311	23	lovatto	lovatto	PROPN
bracis-19021	311	24	,	,	PUNCT
bracis-19021	311	25	 	 	SPACE
bracis-19021	311	26	thiago	thiago	PROPN
bracis-19021	311	27	pereira	pereira	PROPN
bracis-19021	311	28	bueno	bueno	PROPN
bracis-19021	311	29	 	 	SPACE
bracis-19021	311	30	&	&	CCONJ
bracis-19021	311	31	 	 	SPACE
bracis-19021	311	32	leliane	leliane	PROPN
bracis-19021	311	33	nunes	nunes	PROPN
bracis-19021	311	34	de	de	PROPN
bracis-19021	311	35	barros	barros	PROPN
bracis-19021	311	36	authors	author	NOUN
bracis-19021	311	37	ângelo	ângelo	VERB
bracis-19021	311	38	gregório	gregório	PROPN
bracis-19021	311	39	lovattoview	lovattoview	PROPN
bracis-19021	311	40	author	author	NOUN
bracis-19021	311	41	publications	publication	NOUN
bracis-19021	311	42	search	search	NOUN
bracis-19021	311	43	author	author	NOUN
bracis-19021	311	44	on	on	ADP
bracis-19021	311	45	:	:	PUNCT
bracis-19021	311	46	pubmed	pubmed	PROPN
bracis-19021	311	47	 	 	SPACE
bracis-19021	311	48	google	google	PROPN
bracis-19021	311	49	scholar	scholar	PROPN
bracis-19021	311	50	thiago	thiago	PROPN
bracis-19021	311	51	pereira	pereira	PROPN
bracis-19021	311	52	buenoview	buenoview	PROPN
bracis-19021	311	53	author	author	NOUN
bracis-19021	311	54	publications	publication	VERB
bracis-19021	311	55	search	search	NOUN
bracis-19021	311	56	author	author	NOUN
bracis-19021	311	57	on	on	ADP
bracis-19021	311	58	:	:	PUNCT
bracis-19021	311	59	pubmed	pubmed	PROPN
bracis-19021	311	60	 	 	SPACE
bracis-19021	311	61	google	google	PROPN
bracis-19021	311	62	scholar	scholar	NOUN
bracis-19021	311	63	leliane	leliane	PROPN
bracis-19021	311	64	nunes	nunes	PROPN
bracis-19021	311	65	de	de	X
bracis-19021	311	66	barrosview	barrosview	NOUN
bracis-19021	311	67	author	author	NOUN
bracis-19021	311	68	publications	publication	NOUN
bracis-19021	311	69	search	search	NOUN
bracis-19021	311	70	author	author	NOUN
bracis-19021	311	71	on	on	ADP
bracis-19021	311	72	:	:	PUNCT
bracis-19021	311	73	pubmed	pubmed	PROPN
bracis-19021	311	74	 	 	SPACE
bracis-19021	311	75	google	google	PROPN
bracis-19021	311	76	scholar	scholar	NOUN
bracis-19021	311	77	corresponding	correspond	VERB
bracis-19021	311	78	author	author	NOUN
bracis-19021	311	79	correspondence	correspondence	NOUN
bracis-19021	311	80	to	to	ADP
bracis-19021	311	81	ângelo	ângelo	VERB
bracis-19021	311	82	gregório	gregório	PROPN
bracis-19021	311	83	lovatto	lovatto	PROPN
bracis-19021	311	84	.	.	PUNCT
bracis-19021	312	1	editor	editor	NOUN
bracis-19021	312	2	information	information	NOUN
bracis-19021	312	3	editors	editor	NOUN
bracis-19021	312	4	and	and	CCONJ
bracis-19021	312	5	affiliations	affiliation	NOUN
bracis-19021	312	6	universidade	universidade	PROPN
bracis-19021	312	7	federal	federal	PROPN
bracis-19021	312	8	de	de	X
bracis-19021	312	9	sergipe	sergipe	PROPN
bracis-19021	312	10	,	,	PUNCT
bracis-19021	312	11	são	são	NOUN
bracis-19021	312	12	cristóvão	cristóvão	PROPN
bracis-19021	312	13	,	,	PUNCT
bracis-19021	312	14	brazil	brazil	PROPN
bracis-19021	312	15	andré	andré	PROPN
bracis-19021	312	16	britto	britto	PROPN
bracis-19021	312	17	universidade	universidade	PROPN
bracis-19021	312	18	de	de	PROPN
bracis-19021	312	19	são	são	PROPN
bracis-19021	312	20	paulo	paulo	PROPN
bracis-19021	312	21	,	,	PUNCT
bracis-19021	312	22	são	são	PROPN
bracis-19021	312	23	paulo	paulo	PROPN
bracis-19021	312	24	,	,	PUNCT
bracis-19021	312	25	brazil	brazil	PROPN
bracis-19021	312	26	karina	karina	PROPN
bracis-19021	312	27	valdivia	valdivia	PROPN
bracis-19021	312	28	delgado	delgado	VERB
bracis-19021	312	29	rights	right	NOUN
bracis-19021	312	30	and	and	CCONJ
bracis-19021	312	31	permissions	permission	NOUN
bracis-19021	312	32	reprints	reprint	NOUN
bracis-19021	312	33	and	and	CCONJ
bracis-19021	312	34	permissions	permission	VERB
bracis-19021	312	35	copyright	copyright	NOUN
bracis-19021	312	36	information	information	NOUN
bracis-19021	312	37	©	©	PROPN
bracis-19021	312	38	2021	2021	NUM
bracis-19021	312	39	springer	springer	NOUN
bracis-19021	312	40	nature	nature	NOUN
bracis-19021	312	41	switzerland	switzerland	PROPN
bracis-19021	312	42	ag	ag	PROPN
bracis-19021	312	43	about	about	ADP
bracis-19021	312	44	this	this	DET
bracis-19021	312	45	paper	paper	NOUN
bracis-19021	312	46	cite	cite	VERB
bracis-19021	312	47	this	this	DET
bracis-19021	312	48	paper	paper	NOUN
bracis-19021	312	49	lovatto	lovatto	NOUN
bracis-19021	312	50	,	,	PUNCT
bracis-19021	312	51	â.g	â.g	PROPN
bracis-19021	312	52	.	.	PROPN
bracis-19021	312	53	,	,	PUNCT
bracis-19021	312	54	bueno	bueno	PROPN
bracis-19021	312	55	,	,	PUNCT
bracis-19021	312	56	t.p	t.p	PROPN
bracis-19021	312	57	.	.	PROPN
bracis-19021	312	58	,	,	PUNCT
bracis-19021	312	59	de	de	X
bracis-19021	312	60	barros	barros	PROPN
bracis-19021	312	61	,	,	PUNCT
bracis-19021	312	62	l.n	l.n	PROPN
bracis-19021	312	63	.	.	PROPN
bracis-19021	312	64	(	(	PUNCT
bracis-19021	312	65	2021	2021	NUM
bracis-19021	312	66	)	)	PUNCT
bracis-19021	312	67	.	.	PUNCT
bracis-19021	313	1	gradient	gradient	ADJ
bracis-19021	313	2	estimation	estimation	NOUN
bracis-19021	313	3	in	in	ADP
bracis-19021	313	4	model	model	NOUN
bracis-19021	313	5	-	-	PUNCT
bracis-19021	313	6	based	base	VERB
bracis-19021	313	7	reinforcement	reinforcement	NOUN
bracis-19021	313	8	learning	learning	NOUN
bracis-19021	313	9	:	:	PUNCT
bracis-19021	313	10	a	a	DET
bracis-19021	313	11	study	study	NOUN
bracis-19021	313	12	on	on	ADP
bracis-19021	313	13	linear	linear	ADJ
bracis-19021	313	14	quadratic	quadratic	ADJ
bracis-19021	313	15	environments	environment	NOUN
bracis-19021	313	16	.	.	PUNCT
bracis-19021	314	1	in	in	ADP
bracis-19021	314	2	:	:	PUNCT
bracis-19021	314	3	britto	britto	PROPN
bracis-19021	314	4	,	,	PUNCT
bracis-19021	314	5	a.	a.	PROPN
bracis-19021	314	6	,	,	PUNCT
bracis-19021	314	7	valdivia	valdivia	PROPN
bracis-19021	314	8	delgado	delgado	PROPN
bracis-19021	314	9	,	,	PUNCT
bracis-19021	314	10	k.	k.	PROPN
bracis-19021	314	11	(	(	PUNCT
bracis-19021	314	12	eds	eds	PROPN
bracis-19021	314	13	)	)	PUNCT
bracis-19021	314	14	intelligent	intelligent	ADJ
bracis-19021	314	15	systems	system	NOUN
bracis-19021	314	16	.	.	PUNCT
bracis-19021	315	1	bracis	bracis	PROPN
bracis-19021	315	2	2021	2021	NUM
bracis-19021	315	3	.	.	PUNCT
bracis-19021	316	1	lecture	lecture	NOUN
bracis-19021	316	2	notes	note	NOUN
bracis-19021	316	3	in	in	ADP
bracis-19021	316	4	computer	computer	NOUN
bracis-19021	316	5	science	science	NOUN
bracis-19021	316	6	(	(	PUNCT
bracis-19021	316	7	)	)	PUNCT
bracis-19021	316	8	,	,	PUNCT
bracis-19021	316	9	vol	vol	NOUN
bracis-19021	316	10	13073	13073	NUM
bracis-19021	316	11	.	.	PUNCT
bracis-19021	317	1	springer	springer	NOUN
bracis-19021	317	2	,	,	PUNCT
bracis-19021	317	3	cham	cham	PROPN
bracis-19021	317	4	.	.	PUNCT
bracis-19021	318	1	https://doi.org/10.1007/978-3-030-91702-9_3	https://doi.org/10.1007/978-3-030-91702-9_3	PROPN
bracis-19021	318	2	download	download	PROPN
bracis-19021	318	3	citation	citation	NOUN
bracis-19021	318	4	.ris	.ris	PUNCT
bracis-19021	319	1	.enw	.enw	PROPN
bracis-19021	319	2	.bib	.bib	PUNCT
bracis-19021	320	1	doi	doi	PROPN
bracis-19021	320	2	:	:	PUNCT
bracis-19021	320	3	https://doi.org/10.1007/978-3-030-91702-9_3	https://doi.org/10.1007/978-3-030-91702-9_3	PUNCT
bracis-19021	320	4	published	publish	VERB
bracis-19021	320	5	:	:	PUNCT
bracis-19021	320	6	28	28	NUM
bracis-19021	320	7	november	november	PROPN
bracis-19021	320	8	2021	2021	NUM
bracis-19021	320	9	publisher	publisher	NOUN
bracis-19021	320	10	name	name	NOUN
bracis-19021	320	11	:	:	PUNCT
bracis-19021	320	12	springer	springer	NOUN
bracis-19021	320	13	,	,	PUNCT
bracis-19021	320	14	cham	cham	PROPN
bracis-19021	320	15	print	print	PROPN
bracis-19021	320	16	isbn	isbn	PROPN
bracis-19021	320	17	:	:	PUNCT
bracis-19021	320	18	978	978	NUM
bracis-19021	320	19	-	-	SYM
bracis-19021	320	20	3	3	NUM
bracis-19021	320	21	-	-	PUNCT
bracis-19021	320	22	030	030	NUM
bracis-19021	320	23	-	-	PUNCT
bracis-19021	320	24	91701	91701	NUM
bracis-19021	320	25	-	-	SYM
bracis-19021	320	26	2	2	NUM
bracis-19021	320	27	online	online	ADJ
bracis-19021	320	28	isbn	isbn	NOUN
bracis-19021	320	29	:	:	PUNCT
bracis-19021	320	30	978	978	NUM
bracis-19021	320	31	-	-	SYM
bracis-19021	320	32	3	3	NUM
bracis-19021	320	33	-	-	PUNCT
bracis-19021	320	34	030	030	NUM
bracis-19021	320	35	-	-	PUNCT
bracis-19021	320	36	91702	91702	NUM
bracis-19021	320	37	-	-	SYM
bracis-19021	320	38	9	9	NUM
bracis-19021	320	39	ebook	ebook	NOUN
bracis-19021	320	40	packages	package	NOUN
bracis-19021	320	41	:	:	PUNCT
bracis-19021	320	42	computer	computer	NOUN
bracis-19021	320	43	sciencecomputer	sciencecomputer	NOUN
bracis-19021	320	44	science	science	NOUN
bracis-19021	320	45	(	(	PUNCT
bracis-19021	320	46	r0	r0	NOUN
bracis-19021	320	47	)	)	PUNCT
bracis-19021	320	48	share	share	VERB
bracis-19021	320	49	this	this	DET
bracis-19021	320	50	paper	paper	NOUN
bracis-19021	320	51	anyone	anyone	PRON
bracis-19021	320	52	you	you	PRON
bracis-19021	320	53	share	share	VERB
bracis-19021	320	54	the	the	DET
bracis-19021	320	55	following	follow	VERB
bracis-19021	320	56	link	link	NOUN
bracis-19021	320	57	with	with	ADP
bracis-19021	320	58	will	will	AUX
bracis-19021	320	59	be	be	AUX
bracis-19021	320	60	able	able	ADJ
bracis-19021	320	61	to	to	PART
bracis-19021	320	62	read	read	VERB
bracis-19021	320	63	this	this	DET
bracis-19021	320	64	content	content	NOUN
bracis-19021	320	65	:	:	PUNCT
bracis-19021	320	66	get	get	VERB
bracis-19021	320	67	shareable	shareable	ADJ
bracis-19021	320	68	linksorry	linksorry	NOUN
bracis-19021	320	69	,	,	PUNCT
bracis-19021	320	70	a	a	DET
bracis-19021	320	71	shareable	shareable	ADJ
bracis-19021	320	72	link	link	NOUN
bracis-19021	320	73	is	be	AUX
bracis-19021	320	74	not	not	PART
bracis-19021	320	75	currently	currently	ADV
bracis-19021	320	76	available	available	ADJ
bracis-19021	320	77	for	for	ADP
bracis-19021	320	78	this	this	DET
bracis-19021	320	79	article	article	NOUN
bracis-19021	320	80	.	.	PUNCT
bracis-19021	321	1	copy	copy	VERB
bracis-19021	321	2	shareable	shareable	ADJ
bracis-19021	321	3	link	link	NOUN
bracis-19021	321	4	to	to	PART
bracis-19021	321	5	clipboard	clipboard	NOUN
bracis-19021	321	6	provided	provide	VERB
bracis-19021	321	7	by	by	ADP
bracis-19021	321	8	the	the	DET
bracis-19021	321	9	springer	springer	NOUN
bracis-19021	321	10	nature	nature	PROPN
bracis-19021	321	11	sharedit	sharedit	PROPN
bracis-19021	321	12	content	content	NOUN
bracis-19021	321	13	-	-	PUNCT
bracis-19021	321	14	sharing	share	VERB
bracis-19021	321	15	initiative	initiative	NOUN
bracis-19021	321	16	keywords	keyword	NOUN
bracis-19021	321	17	reinforcement	reinforcement	NOUN
bracis-19021	321	18	learning	learn	VERB
bracis-19021	321	19	model	model	NOUN
bracis-19021	321	20	-	-	PUNCT
bracis-19021	321	21	based	base	VERB
bracis-19021	321	22	machine	machine	NOUN
bracis-19021	321	23	learning	learning	NOUN
bracis-19021	321	24	publish	publish	VERB
bracis-19021	321	25	with	with	ADP
bracis-19021	321	26	us	us	PROPN
bracis-19021	321	27	policies	policy	NOUN
bracis-19021	321	28	and	and	CCONJ
bracis-19021	321	29	ethics	ethic	NOUN
bracis-19021	321	30	search	search	NOUN
bracis-19021	321	31	search	search	NOUN
bracis-19021	321	32	by	by	ADP
bracis-19021	321	33	keyword	keyword	NOUN
bracis-19021	321	34	or	or	CCONJ
bracis-19021	321	35	author	author	NOUN
bracis-19021	321	36	search	search	NOUN
bracis-19021	321	37	navigation	navigation	NOUN
bracis-19021	321	38	find	find	VERB
bracis-19021	321	39	a	a	DET
bracis-19021	321	40	journal	journal	NOUN
bracis-19021	321	41	publish	publish	VERB
bracis-19021	321	42	with	with	ADP
bracis-19021	321	43	us	we	PRON
bracis-19021	321	44	track	track	VERB
bracis-19021	321	45	your	your	PRON
bracis-19021	321	46	research	research	NOUN
bracis-19021	321	47	discover	discover	VERB
bracis-19021	321	48	content	content	NOUN
bracis-19021	321	49	journals	journal	NOUN
bracis-19021	321	50	a	a	DET
bracis-19021	321	51	-	-	PUNCT
bracis-19021	321	52	z	z	NOUN
bracis-19021	321	53	books	book	NOUN
bracis-19021	321	54	a	a	DET
bracis-19021	321	55	-	-	PUNCT
bracis-19021	321	56	z	z	NOUN
bracis-19021	321	57	publish	publish	NOUN
bracis-19021	321	58	with	with	ADP
bracis-19021	321	59	us	us	PROPN
bracis-19021	321	60	journal	journal	PROPN
bracis-19021	321	61	finder	finder	PROPN
bracis-19021	321	62	publish	publish	VERB
bracis-19021	321	63	your	your	PRON
bracis-19021	321	64	research	research	NOUN
bracis-19021	321	65	language	language	NOUN
bracis-19021	321	66	editing	edit	VERB
bracis-19021	321	67	open	open	ADJ
bracis-19021	321	68	access	access	NOUN
bracis-19021	321	69	publishing	publishing	NOUN
bracis-19021	321	70	products	product	NOUN
bracis-19021	321	71	and	and	CCONJ
bracis-19021	321	72	services	service	NOUN
bracis-19021	321	73	our	our	PRON
bracis-19021	321	74	products	product	NOUN
bracis-19021	321	75	librarians	librarian	VERB
bracis-19021	321	76	societies	society	NOUN
bracis-19021	321	77	partners	partner	NOUN
bracis-19021	321	78	and	and	CCONJ
bracis-19021	321	79	advertisers	advertiser	NOUN
bracis-19021	321	80	our	our	PRON
bracis-19021	321	81	brands	brand	NOUN
bracis-19021	321	82	springer	springer	NOUN
bracis-19021	321	83	nature	nature	PROPN
bracis-19021	321	84	portfolio	portfolio	PROPN
bracis-19021	321	85	bmc	bmc	PROPN
bracis-19021	321	86	palgrave	palgrave	PROPN
bracis-19021	321	87	macmillan	macmillan	PROPN
bracis-19021	321	88	apress	apress	PROPN
bracis-19021	321	89	discover	discover	VERB
bracis-19021	321	90	your	your	PRON
bracis-19021	321	91	privacy	privacy	NOUN
bracis-19021	321	92	choices	choice	NOUN
bracis-19021	321	93	/	/	SYM
bracis-19021	321	94	manage	manage	NOUN
bracis-19021	321	95	cookies	cookie	NOUN
bracis-19021	321	96	your	your	PRON
bracis-19021	321	97	us	us	PROPN
bracis-19021	322	1	state	state	NOUN
bracis-19021	322	2	privacy	privacy	NOUN
bracis-19021	322	3	rights	right	NOUN
bracis-19021	322	4	accessibility	accessibility	NOUN
bracis-19021	322	5	statement	statement	NOUN
bracis-19021	322	6	terms	term	NOUN
bracis-19021	322	7	and	and	CCONJ
bracis-19021	322	8	conditions	condition	NOUN
bracis-19021	322	9	privacy	privacy	NOUN
bracis-19021	322	10	policy	policy	NOUN
bracis-19021	322	11	help	help	NOUN
bracis-19021	322	12	and	and	CCONJ
bracis-19021	322	13	support	support	VERB
bracis-19021	322	14	legal	legal	ADJ
bracis-19021	322	15	notice	notice	NOUN
bracis-19021	322	16	cancel	cancel	VERB
bracis-19021	322	17	contracts	contract	NOUN
bracis-19021	322	18	here	here	ADV
bracis-19021	322	19	129.74.145.123	129.74.145.123	NUM
bracis-19021	322	20	hesburgh	hesburgh	PROPN
bracis-19021	322	21	library	library	PROPN
bracis-19021	322	22	er	er	INTJ
bracis-19021	322	23	unit	unit	NOUN
bracis-19021	322	24	(	(	PUNCT
bracis-19021	322	25	3005732405	3005732405	NUM
bracis-19021	322	26	)	)	PUNCT
bracis-19021	322	27	northeast	northeast	ADJ
bracis-19021	322	28	research	research	NOUN
bracis-19021	322	29	libraries	library	NOUN
bracis-19021	322	30	(	(	PUNCT
bracis-19021	322	31	nerl	nerl	PROPN
bracis-19021	322	32	)	)	PUNCT
bracis-19021	322	33	(	(	PUNCT
bracis-19021	322	34	8200828607	8200828607	NUM
bracis-19021	322	35	)	)	PUNCT
bracis-19021	322	36	nerl	nerl	VERB
bracis-19021	322	37	ta	ta	X
bracis-19021	322	38	account	account	NOUN
bracis-19021	322	39	(	(	PUNCT
bracis-19021	322	40	3006206169	3006206169	NUM
bracis-19021	322	41	)	)	PUNCT
bracis-19021	322	42	university	university	NOUN
bracis-19021	322	43	of	of	ADP
bracis-19021	322	44	notre	notre	PROPN
bracis-19021	322	45	dame	dame	PROPN
bracis-19021	322	46	hesburgh	hesburgh	PROPN
bracis-19021	322	47	library	library	NOUN
bracis-19021	322	48	(	(	PUNCT
bracis-19021	322	49	3000184373	3000184373	NUM
bracis-19021	322	50	)	)	PUNCT
bracis-19021	323	1	©	©	ADP
bracis-19021	323	2	2025	2025	NUM
bracis-19021	323	3	springer	springer	NOUN
bracis-19021	323	4	nature	nature	NOUN