id	sid	tid	token	lemma	pos
cana-2550	1	1	communications	communication	NOUN
cana-2550	1	2	on	on	ADP
cana-2550	1	3	applied	apply	VERB
cana-2550	1	4	nonlinear	nonlinear	ADJ
cana-2550	1	5	analysis	analysis	NOUN
cana-2550	1	6	issn	issn	NOUN
cana-2550	1	7	:	:	PUNCT
cana-2550	1	8	1074	1074	NUM
cana-2550	1	9	-	-	PUNCT
cana-2550	1	10	133x	133x	NUM
cana-2550	1	11	vol	vol	NOUN
cana-2550	1	12	32	32	NUM
cana-2550	1	13	no	no	NOUN
cana-2550	1	14	.	.	PUNCT
cana-2550	2	1	3s	3s	NUM
cana-2550	2	2	(	(	PUNCT
cana-2550	2	3	2025	2025	NUM
cana-2550	2	4	)	)	PUNCT
cana-2550	2	5	68	68	NUM
cana-2550	2	6	https://internationalpubls.com	https://internationalpubls.com	X
cana-2550	2	7	optimization	optimization	NOUN
cana-2550	2	8	in	in	ADP
cana-2550	2	9	neural	neural	ADJ
cana-2550	2	10	networks	network	NOUN
cana-2550	2	11	with	with	ADP
cana-2550	2	12	mathematics	mathematic	NOUN
cana-2550	2	13	:	:	PUNCT
cana-2550	2	14	efficiency	efficiency	NOUN
cana-2550	2	15	in	in	ADP
cana-2550	2	16	architectures	architecture	NOUN
cana-2550	2	17	of	of	ADP
cana-2550	2	18	deep	deep	ADJ
cana-2550	2	19	learning	learning	NOUN
cana-2550	2	20	1dr	1dr	NOUN
cana-2550	2	21	.	.	PUNCT
cana-2550	3	1	anandhavalli	anandhavalli	PROPN
cana-2550	3	2	muniasamy	muniasamy	ADJ
cana-2550	3	3	,	,	PUNCT
cana-2550	3	4	2dr	2dr	ADJ
cana-2550	3	5	.	.	PUNCT
cana-2550	4	1	sanjaykumar	sanjaykumar	PROPN
cana-2550	4	2	p.	p.	PROPN
cana-2550	4	3	pingat	pingat	PROPN
cana-2550	4	4	,	,	PUNCT
cana-2550	4	5	3harshitha	3harshitha	NUM
cana-2550	4	6	raghavan	raghavan	PROPN
cana-2550	4	7	devarajan	devarajan	PROPN
cana-2550	4	8	,	,	PUNCT
cana-2550	4	9	4dr	4dr	ADJ
cana-2550	4	10	jayasundar	jayasundar	NOUN
cana-2550	4	11	s	s	PART
cana-2550	4	12	,	,	PUNCT
cana-2550	4	13	5vikas	5vikas	PROPN
cana-2550	4	14	kumar	kumar	PROPN
cana-2550	4	15	,	,	PUNCT
cana-2550	4	16	6shailesh	6shailesh	PROPN
cana-2550	4	17	v	v	ADP
cana-2550	4	18	kulkarni	kulkarni	PROPN
cana-2550	4	19	,	,	PUNCT
cana-2550	4	20	7dr	7dr	ADJ
cana-2550	4	21	.	.	PUNCT
cana-2550	5	1	soibam	soibam	PROPN
cana-2550	5	2	birajit	birajit	PROPN
cana-2550	5	3	singh	singh	PROPN
cana-2550	5	4	.	.	PUNCT
cana-2550	6	1	1associate	1associate	NUM
cana-2550	6	2	professor	professor	NOUN
cana-2550	6	3	,	,	PUNCT
cana-2550	6	4	department	department	NOUN
cana-2550	6	5	of	of	ADP
cana-2550	6	6	informatics	informatic	NOUN
cana-2550	6	7	and	and	CCONJ
cana-2550	6	8	computer	computer	NOUN
cana-2550	6	9	systems	system	NOUN
cana-2550	6	10	,	,	PUNCT
cana-2550	6	11	college	college	NOUN
cana-2550	6	12	of	of	ADP
cana-2550	6	13	computer	computer	NOUN
cana-2550	6	14	science	science	NOUN
cana-2550	6	15	,	,	PUNCT
cana-2550	6	16	king	king	PROPN
cana-2550	6	17	khalid	khalid	PROPN
cana-2550	6	18	university	university	PROPN
cana-2550	6	19	,	,	PUNCT
cana-2550	6	20	saudi	saudi	PROPN
cana-2550	6	21	arabia	arabia	PROPN
cana-2550	6	22	.	.	PUNCT
cana-2550	7	1	email	email	NOUN
cana-2550	7	2	:	:	PUNCT
cana-2550	7	3	anandhavalli.dr@gmail.com	anandhavalli.dr@gmail.com	PROPN
cana-2550	7	4	.	.	PUNCT
cana-2550	8	1	orcid	orcid	NOUN
cana-2550	8	2	:	:	PUNCT
cana-2550	8	3	0000	0000	NUM
cana-2550	8	4	-	-	PUNCT
cana-2550	8	5	0001	0001	NUM
cana-2550	8	6	-	-	PUNCT
cana-2550	8	7	8940	8940	NUM
cana-2550	8	8	-	-	SYM
cana-2550	8	9	3954	3954	NUM
cana-2550	8	10	2assistant	2assistant	PROPN
cana-2550	8	11	professor	professor	NOUN
cana-2550	8	12	,	,	PUNCT
cana-2550	8	13	computer	computer	NOUN
cana-2550	8	14	engineering	engineering	NOUN
cana-2550	8	15	,	,	PUNCT
cana-2550	8	16	smt	smt	PROPN
cana-2550	8	17	.	.	PUNCT
cana-2550	9	1	kashibai	kashibai	PROPN
cana-2550	9	2	navale	navale	PROPN
cana-2550	9	3	college	college	PROPN
cana-2550	9	4	of	of	ADP
cana-2550	9	5	engineering	engineering	NOUN
cana-2550	9	6	.	.	PUNCT
cana-2550	10	1	email	email	NOUN
cana-2550	10	2	:	:	PUNCT
cana-2550	11	1	sanjaypingat@gmail.com	sanjaypingat@gmail.com	X
cana-2550	12	1	3data	3data	NUM
cana-2550	12	2	scientist	scientist	NOUN
cana-2550	12	3	,	,	PUNCT
cana-2550	12	4	amazon	amazon	PROPN
cana-2550	12	5	,	,	PUNCT
cana-2550	12	6	email	email	NOUN
cana-2550	12	7	i	i	PROPN
cana-2550	12	8	d	d	PROPN
cana-2550	12	9	:	:	PUNCT
cana-2550	12	10	harshithard05@gmail.com	harshithard05@gmail.com	NOUN
cana-2550	12	11	,	,	PUNCT
cana-2550	12	12	orcid	orcid	PROPN
cana-2550	12	13	:	:	PUNCT
cana-2550	12	14	0000	0000	NUM
cana-2550	12	15	-	-	PUNCT
cana-2550	12	16	0003	0003	NUM
cana-2550	12	17	-	-	PUNCT
cana-2550	12	18	0330	0330	NUM
cana-2550	12	19	-	-	SYM
cana-2550	12	20	9698	9698	NUM
cana-2550	12	21	4professor	4professor	NUM
cana-2550	12	22	,	,	PUNCT
cana-2550	12	23	computer	computer	NOUN
cana-2550	12	24	science	science	NOUN
cana-2550	12	25	and	and	CCONJ
cana-2550	12	26	engineering	engineering	NOUN
cana-2550	12	27	,	,	PUNCT
cana-2550	12	28	idhaya	idhaya	NOUN
cana-2550	12	29	engineering	engineering	NOUN
cana-2550	12	30	college	college	NOUN
cana-2550	12	31	for	for	ADP
cana-2550	12	32	women	woman	NOUN
cana-2550	12	33	,	,	PUNCT
cana-2550	12	34	chinnasalem	chinnasalem	NOUN
cana-2550	12	35	,	,	PUNCT
cana-2550	12	36	tamil	tamil	PROPN
cana-2550	12	37	nadu	nadu	NOUN
cana-2550	12	38	,	,	PUNCT
cana-2550	12	39	pin	pin	NOUN
cana-2550	12	40	code-606201	code-606201	NOUN
cana-2550	12	41	,	,	PUNCT
cana-2550	12	42	india	india	PROPN
cana-2550	12	43	.	.	PUNCT
cana-2550	13	1	e	e	X
cana-2550	13	2	-	-	NOUN
cana-2550	13	3	mail	mail	NOUN
cana-2550	13	4	id:chisundar123@gmail.com	id:chisundar123@gmail.com	ADJ
cana-2550	13	5	.	.	PUNCT
cana-2550	14	1	orcid	orcid	NOUN
cana-2550	14	2	:	:	PUNCT
cana-2550	14	3	0000	0000	NUM
cana-2550	14	4	-	-	PUNCT
cana-2550	14	5	0002	0002	NUM
cana-2550	14	6	-	-	PUNCT
cana-2550	14	7	6456	6456	NUM
cana-2550	14	8	-	-	SYM
cana-2550	14	9	9277	9277	NUM
cana-2550	14	10	5assistant	5assistant	NUM
cana-2550	14	11	professor	professor	NOUN
cana-2550	14	12	,	,	PUNCT
cana-2550	14	13	school	school	NOUN
cana-2550	14	14	of	of	ADP
cana-2550	14	15	applied	applied	ADJ
cana-2550	14	16	and	and	CCONJ
cana-2550	14	17	life	life	NOUN
cana-2550	14	18	sciences	science	NOUN
cana-2550	14	19	,	,	PUNCT
cana-2550	14	20	uttaranchal	uttaranchal	PROPN
cana-2550	14	21	university	university	NOUN
cana-2550	14	22	,	,	PUNCT
cana-2550	14	23	dehradun	dehradun	PROPN
cana-2550	14	24	,	,	PUNCT
cana-2550	14	25	248007	248007	NUM
cana-2550	14	26	,	,	PUNCT
cana-2550	15	1	india	india	PROPN
cana-2550	15	2	.	.	PUNCT
cana-2550	15	3	email	email	NOUN
cana-2550	15	4	:	:	PUNCT
cana-2550	15	5	vikas.mathematica@gmail.com	vikas.mathematica@gmail.com	X
cana-2550	15	6	.	.	PUNCT
cana-2550	16	1	orchid	orchid	NOUN
cana-2550	16	2	:	:	PUNCT
cana-2550	16	3	0000	0000	NUM
cana-2550	16	4	-	-	SYM
cana-2550	16	5	0003	0003	NUM
cana-2550	16	6	-	-	PUNCT
cana-2550	16	7	3170	3170	NUM
cana-2550	16	8	-	-	SYM
cana-2550	16	9	7933	7933	NUM
cana-2550	16	10	6professor	6professor	NUM
cana-2550	16	11	,	,	PUNCT
cana-2550	16	12	department	department	NOUN
cana-2550	16	13	of	of	ADP
cana-2550	16	14	electronics	electronic	NOUN
cana-2550	16	15	and	and	CCONJ
cana-2550	16	16	telecommunication	telecommunication	NOUN
cana-2550	16	17	.	.	PUNCT
cana-2550	17	1	vishwakarma	vishwakarma	PROPN
cana-2550	17	2	institute	institute	PROPN
cana-2550	17	3	of	of	ADP
cana-2550	17	4	technology	technology	PROPN
cana-2550	17	5	,	,	PUNCT
cana-2550	17	6	pune-37	pune-37	PROPN
cana-2550	17	7	.	.	PUNCT
cana-2550	17	8	shailesh.kulkarni@vit.edu	shailesh.kulkarni@vit.edu	PROPN
cana-2550	17	9	.	.	PUNCT
cana-2550	18	1	orcid	orcid	NOUN
cana-2550	18	2	:	:	PUNCT
cana-2550	18	3	0000	0000	NUM
cana-2550	18	4	-	-	PUNCT
cana-2550	18	5	0003	0003	NUM
cana-2550	18	6	-	-	PUNCT
cana-2550	18	7	3764	3764	NUM
cana-2550	18	8	-	-	PUNCT
cana-2550	18	9	8890	8890	NUM
cana-2550	18	10	7assistant	7assistant	ADJ
cana-2550	18	11	professor	professor	NOUN
cana-2550	18	12	(	(	PUNCT
cana-2550	18	13	education	education	NOUN
cana-2550	18	14	)	)	PUNCT
cana-2550	18	15	,	,	PUNCT
cana-2550	18	16	manipur	manipur	PROPN
cana-2550	18	17	college	college	PROPN
cana-2550	18	18	,	,	PUNCT
cana-2550	18	19	imphal	imphal	PROPN
cana-2550	18	20	.	.	PUNCT
cana-2550	19	1	orcid	orcid	NOUN
cana-2550	19	2	:	:	PUNCT
cana-2550	19	3	0009	0009	NUM
cana-2550	19	4	-	-	PUNCT
cana-2550	19	5	0005	0005	NUM
cana-2550	19	6	-	-	PUNCT
cana-2550	19	7	8911	8911	NUM
cana-2550	19	8	-	-	SYM
cana-2550	19	9	5713	5713	NUM
cana-2550	19	10	.	.	PUNCT
cana-2550	20	1	email	email	NOUN
cana-2550	20	2	:	:	PUNCT
cana-2550	20	3	birajits@manipurcollege.ac.in	birajits@manipurcollege.ac.in	NOUN
cana-2550	20	4	article	article	NOUN
cana-2550	20	5	history	history	NOUN
cana-2550	20	6	:	:	PUNCT
cana-2550	20	7	received	receive	VERB
cana-2550	20	8	:	:	PUNCT
cana-2550	20	9	20	20	NUM
cana-2550	20	10	-	-	SYM
cana-2550	20	11	09	09	NUM
cana-2550	20	12	-	-	PUNCT
cana-2550	20	13	2024	2024	NUM
cana-2550	20	14	revised	revise	VERB
cana-2550	20	15	:	:	PUNCT
cana-2550	20	16	03	03	NUM
cana-2550	20	17	-	-	SYM
cana-2550	20	18	11	11	NUM
cana-2550	20	19	-	-	PUNCT
cana-2550	20	20	2024	2024	NUM
cana-2550	20	21	accepted	accept	VERB
cana-2550	20	22	:	:	PUNCT
cana-2550	20	23	17	17	NUM
cana-2550	20	24	-	-	SYM
cana-2550	20	25	11	11	NUM
cana-2550	20	26	-	-	PUNCT
cana-2550	20	27	2024	2024	NUM
cana-2550	20	28	abstract	abstract	NOUN
cana-2550	20	29	:	:	PUNCT
cana-2550	20	30	optimization	optimization	NOUN
cana-2550	20	31	is	be	AUX
cana-2550	20	32	essential	essential	ADJ
cana-2550	20	33	for	for	ADP
cana-2550	20	34	improving	improve	VERB
cana-2550	20	35	the	the	DET
cana-2550	20	36	effectiveness	effectiveness	NOUN
cana-2550	20	37	and	and	CCONJ
cana-2550	20	38	performance	performance	NOUN
cana-2550	20	39	of	of	ADP
cana-2550	20	40	neural	neural	ADJ
cana-2550	20	41	networks	network	NOUN
cana-2550	20	42	specially	specially	ADV
cana-2550	20	43	in	in	ADP
cana-2550	20	44	deep	deep	ADJ
cana-2550	20	45	learning	learning	NOUN
cana-2550	20	46	.	.	PUNCT
cana-2550	21	1	in	in	ADP
cana-2550	21	2	this	this	DET
cana-2550	21	3	paper	paper	NOUN
cana-2550	21	4	the	the	DET
cana-2550	21	5	concepts	concept	NOUN
cana-2550	21	6	of	of	ADP
cana-2550	21	7	mathematical	mathematical	ADJ
cana-2550	21	8	optimization	optimization	NOUN
cana-2550	21	9	methods	method	NOUN
cana-2550	21	10	are	be	AUX
cana-2550	21	11	discussed	discuss	VERB
cana-2550	21	12	that	that	PRON
cana-2550	21	13	improves	improve	VERB
cana-2550	21	14	generalization	generalization	NOUN
cana-2550	21	15	,	,	PUNCT
cana-2550	21	16	convergence	convergence	NOUN
cana-2550	21	17	speed	speed	NOUN
cana-2550	21	18	,	,	PUNCT
cana-2550	21	19	and	and	CCONJ
cana-2550	21	20	model	model	NOUN
cana-2550	21	21	training	training	NOUN
cana-2550	21	22	.	.	PUNCT
cana-2550	22	1	the	the	DET
cana-2550	22	2	paper	paper	NOUN
cana-2550	22	3	will	will	AUX
cana-2550	22	4	discuss	discuss	VERB
cana-2550	22	5	both	both	DET
cana-2550	22	6	traditional	traditional	ADJ
cana-2550	22	7	and	and	CCONJ
cana-2550	22	8	contemporary	contemporary	ADJ
cana-2550	22	9	techniques	technique	NOUN
cana-2550	22	10	,	,	PUNCT
cana-2550	22	11	such	such	ADJ
cana-2550	22	12	as	as	ADP
cana-2550	22	13	gradient	gradient	NOUN
cana-2550	22	14	-	-	PUNCT
cana-2550	22	15	based	base	VERB
cana-2550	22	16	methods	method	NOUN
cana-2550	22	17	,	,	PUNCT
cana-2550	22	18	second	second	ADJ
cana-2550	22	19	-	-	PUNCT
cana-2550	22	20	order	order	NOUN
cana-2550	22	21	optimization	optimization	NOUN
cana-2550	22	22	,	,	PUNCT
cana-2550	22	23	and	and	CCONJ
cana-2550	22	24	new	new	ADJ
cana-2550	22	25	developments	development	NOUN
cana-2550	22	26	like	like	ADP
cana-2550	22	27	meta	meta	ADJ
cana-2550	22	28	-	-	PUNCT
cana-2550	22	29	optimization	optimization	NOUN
cana-2550	22	30	and	and	CCONJ
cana-2550	22	31	adaptive	adaptive	ADJ
cana-2550	22	32	learning	learning	NOUN
cana-2550	22	33	rate	rate	NOUN
cana-2550	22	34	strategies	strategy	NOUN
cana-2550	22	35	.	.	PUNCT
cana-2550	23	1	there	there	PRON
cana-2550	23	2	are	be	VERB
cana-2550	23	3	some	some	DET
cana-2550	23	4	mathematical	mathematical	ADJ
cana-2550	23	5	formulations	formulation	NOUN
cana-2550	23	6	that	that	PRON
cana-2550	23	7	helps	help	VERB
cana-2550	23	8	to	to	PART
cana-2550	23	9	reduce	reduce	VERB
cana-2550	23	10	loss	loss	NOUN
cana-2550	23	11	functions	function	NOUN
cana-2550	23	12	while	while	SCONJ
cana-2550	23	13	maintaining	maintain	VERB
cana-2550	23	14	the	the	DET
cana-2550	23	15	computational	computational	ADJ
cana-2550	23	16	efficiency	efficiency	NOUN
cana-2550	23	17	.	.	PUNCT
cana-2550	24	1	there	there	PRON
cana-2550	24	2	are	be	VERB
cana-2550	24	3	many	many	ADJ
cana-2550	24	4	optimization	optimization	NOUN
cana-2550	24	5	challenges	challenge	NOUN
cana-2550	24	6	in	in	ADP
cana-2550	24	7	large	large	ADJ
cana-2550	24	8	-	-	PUNCT
cana-2550	24	9	scale	scale	NOUN
cana-2550	24	10	networks	network	NOUN
cana-2550	24	11	which	which	PRON
cana-2550	24	12	will	will	AUX
cana-2550	24	13	be	be	AUX
cana-2550	24	14	discussed	discuss	VERB
cana-2550	24	15	in	in	ADP
cana-2550	24	16	the	the	DET
cana-2550	24	17	paper	paper	NOUN
cana-2550	24	18	and	and	CCONJ
cana-2550	24	19	hence	hence	ADV
cana-2550	24	20	provides	provide	VERB
cana-2550	24	21	innovative	innovative	ADJ
cana-2550	24	22	ways	way	NOUN
cana-2550	24	23	to	to	PART
cana-2550	24	24	strike	strike	VERB
cana-2550	24	25	a	a	DET
cana-2550	24	26	balance	balance	NOUN
cana-2550	24	27	between	between	ADP
cana-2550	24	28	speed	speed	NOUN
cana-2550	24	29	,	,	PUNCT
cana-2550	24	30	accuracy	accuracy	NOUN
cana-2550	24	31	,	,	PUNCT
cana-2550	24	32	and	and	CCONJ
cana-2550	24	33	resource	resource	NOUN
cana-2550	24	34	usage	usage	NOUN
cana-2550	24	35	.	.	PUNCT
cana-2550	25	1	the	the	DET
cana-2550	25	2	paper	paper	NOUN
cana-2550	25	3	also	also	ADV
cana-2550	25	4	discusses	discuss	VERB
cana-2550	25	5	the	the	DET
cana-2550	25	6	role	role	NOUN
cana-2550	25	7	of	of	ADP
cana-2550	25	8	mathematical	mathematical	ADJ
cana-2550	25	9	optimization	optimization	NOUN
cana-2550	25	10	techniques	technique	NOUN
cana-2550	25	11	which	which	PRON
cana-2550	25	12	helps	help	VERB
cana-2550	25	13	to	to	PART
cana-2550	25	14	enhance	enhance	VERB
cana-2550	25	15	the	the	DET
cana-2550	25	16	accuracy	accuracy	NOUN
cana-2550	25	17	,	,	PUNCT
cana-2550	25	18	convergence	convergence	NOUN
cana-2550	25	19	,	,	PUNCT
cana-2550	25	20	speed	speed	NOUN
cana-2550	25	21	,	,	PUNCT
cana-2550	25	22	etc	etc	X
cana-2550	25	23	in	in	ADP
cana-2550	25	24	detail	detail	NOUN
cana-2550	25	25	.	.	PUNCT
cana-2550	26	1	keywords	keyword	NOUN
cana-2550	26	2	:	:	PUNCT
cana-2550	26	3	neural	neural	ADJ
cana-2550	26	4	network	network	NOUN
cana-2550	26	5	optimization	optimization	NOUN
cana-2550	26	6	,	,	PUNCT
cana-2550	26	7	deep	deep	ADJ
cana-2550	26	8	learning	learning	NOUN
cana-2550	26	9	architectures	architecture	NOUN
cana-2550	26	10	,	,	PUNCT
cana-2550	26	11	mathematical	mathematical	ADJ
cana-2550	26	12	optimization	optimization	NOUN
cana-2550	26	13	,	,	PUNCT
cana-2550	26	14	gradient	gradient	NOUN
cana-2550	26	15	-	-	PUNCT
cana-2550	26	16	based	base	VERB
cana-2550	26	17	methods	method	NOUN
cana-2550	26	18	,	,	PUNCT
cana-2550	26	19	adaptive	adaptive	ADJ
cana-2550	26	20	learning	learning	NOUN
cana-2550	26	21	rate	rate	NOUN
cana-2550	26	22	,	,	PUNCT
cana-2550	26	23	convergence	convergence	NOUN
cana-2550	26	24	efficiency	efficiency	NOUN
cana-2550	26	25	,	,	PUNCT
cana-2550	26	26	loss	loss	NOUN
cana-2550	26	27	function	function	NOUN
cana-2550	26	28	minimization	minimization	NOUN
cana-2550	26	29	,	,	PUNCT
cana-2550	26	30	second	second	ADJ
cana-2550	26	31	-	-	PUNCT
cana-2550	26	32	order	order	NOUN
cana-2550	26	33	optimization	optimization	NOUN
cana-2550	26	34	,	,	PUNCT
cana-2550	26	35	meta	meta	NOUN
cana-2550	26	36	-	-	PUNCT
cana-2550	26	37	optimization	optimization	NOUN
cana-2550	26	38	.	.	PUNCT
cana-2550	27	1	1	1	X
cana-2550	27	2	.	.	X
cana-2550	27	3	introduction	introduction	NOUN
cana-2550	27	4	deep	deep	ADJ
cana-2550	27	5	learning	learning	NOUN
cana-2550	27	6	has	have	AUX
cana-2550	27	7	become	become	VERB
cana-2550	27	8	a	a	DET
cana-2550	27	9	revolutionary	revolutionary	ADJ
cana-2550	27	10	technology	technology	NOUN
cana-2550	27	11	in	in	ADP
cana-2550	27	12	the	the	DET
cana-2550	27	13	field	field	NOUN
cana-2550	27	14	of	of	ADP
cana-2550	27	15	computer	computer	NOUN
cana-2550	27	16	vision	vision	NOUN
cana-2550	27	17	,	,	PUNCT
cana-2550	27	18	natural	natural	ADJ
cana-2550	27	19	language	language	NOUN
cana-2550	27	20	processing	processing	NOUN
cana-2550	27	21	,	,	PUNCT
cana-2550	27	22	healthcare	healthcare	NOUN
cana-2550	27	23	,	,	PUNCT
cana-2550	27	24	and	and	CCONJ
cana-2550	27	25	finance	finance	NOUN
cana-2550	27	26	.	.	PUNCT
cana-2550	28	1	the	the	DET
cana-2550	28	2	foundation	foundation	NOUN
cana-2550	28	3	of	of	ADP
cana-2550	28	4	deep	deep	ADJ
cana-2550	28	5	learning	learning	NOUN
cana-2550	28	6	models	model	NOUN
cana-2550	28	7	are	be	AUX
cana-2550	28	8	neural	neural	ADJ
cana-2550	28	9	networks	network	NOUN
cana-2550	28	10	,	,	PUNCT
cana-2550	28	11	which	which	PRON
cana-2550	28	12	consist	consist	VERB
cana-2550	28	13	of	of	ADP
cana-2550	28	14	layers	layer	NOUN
cana-2550	28	15	of	of	ADP
cana-2550	28	16	interconnected	interconnected	ADJ
cana-2550	28	17	neurons	neuron	NOUN
cana-2550	28	18	that	that	PRON
cana-2550	28	19	learn	learn	VERB
cana-2550	28	20	to	to	PART
cana-2550	28	21	map	map	VERB
cana-2550	28	22	inputs	input	NOUN
cana-2550	28	23	to	to	ADP
cana-2550	28	24	outputs	output	NOUN
cana-2550	28	25	through	through	ADP
cana-2550	28	26	optimization	optimization	NOUN
cana-2550	28	27	.	.	PUNCT
cana-2550	29	1	one	one	NUM
cana-2550	29	2	of	of	ADP
cana-2550	29	3	the	the	DET
cana-2550	29	4	biggest	big	ADJ
cana-2550	29	5	success	success	NOUN
cana-2550	29	6	of	of	ADP
cana-2550	29	7	neural	neural	ADJ
cana-2550	29	8	networks	network	NOUN
cana-2550	29	9	has	have	AUX
cana-2550	29	10	been	be	AUX
cana-2550	29	11	the	the	DET
cana-2550	29	12	training	training	NOUN
cana-2550	29	13	and	and	CCONJ
cana-2550	29	14	optimization	optimization	NOUN
cana-2550	29	15	process	process	NOUN
cana-2550	29	16	,	,	PUNCT
cana-2550	29	17	which	which	PRON
cana-2550	29	18	is	be	AUX
cana-2550	29	19	computationally	computationally	ADV
cana-2550	29	20	complex	complex	ADJ
cana-2550	29	21	and	and	CCONJ
cana-2550	29	22	complicated	complicated	ADJ
cana-2550	29	23	and	and	CCONJ
cana-2550	29	24	requires	require	VERB
cana-2550	29	25	significant	significant	ADJ
cana-2550	29	26	amount	amount	NOUN
cana-2550	29	27	of	of	ADP
cana-2550	29	28	resources	resource	NOUN
cana-2550	29	29	and	and	CCONJ
cana-2550	29	30	time	time	NOUN
cana-2550	29	31	.	.	PUNCT
cana-2550	30	1	to	to	PART
cana-2550	30	2	optimize	optimize	VERB
cana-2550	30	3	neural	neural	ADJ
cana-2550	30	4	networks	network	NOUN
cana-2550	30	5	we	we	PRON
cana-2550	30	6	need	need	VERB
cana-2550	30	7	to	to	PART
cana-2550	30	8	minimize	minimize	VERB
cana-2550	30	9	a	a	DET
cana-2550	30	10	loss	loss	NOUN
cana-2550	30	11	function	function	NOUN
cana-2550	30	12	that	that	PRON
cana-2550	30	13	quantifies	quantify	VERB
cana-2550	30	14	the	the	DET
cana-2550	30	15	disparity	disparity	NOUN
cana-2550	30	16	between	between	ADP
cana-2550	30	17	predicted	predict	VERB
cana-2550	30	18	and	and	CCONJ
cana-2550	30	19	actual	actual	ADJ
cana-2550	30	20	outputs	output	NOUN
cana-2550	30	21	.	.	PUNCT
cana-2550	31	1	the	the	DET
cana-2550	31	2	complex	complex	ADJ
cana-2550	31	3	aspects	aspect	NOUN
cana-2550	31	4	of	of	ADP
cana-2550	31	5	this	this	DET
cana-2550	31	6	process	process	NOUN
cana-2550	31	7	are	be	AUX
cana-2550	31	8	navigating	navigate	VERB
cana-2550	31	9	communications	communication	NOUN
cana-2550	31	10	on	on	ADP
cana-2550	31	11	applied	apply	VERB
cana-2550	31	12	nonlinear	nonlinear	ADJ
cana-2550	31	13	analysis	analysis	NOUN
cana-2550	31	14	issn	issn	NOUN
cana-2550	31	15	:	:	PUNCT
cana-2550	31	16	1074	1074	NUM
cana-2550	31	17	-	-	PUNCT
cana-2550	31	18	133x	133x	NUM
cana-2550	31	19	vol	vol	NOUN
cana-2550	31	20	32	32	NUM
cana-2550	31	21	no	no	NOUN
cana-2550	31	22	.	.	PUNCT
cana-2550	32	1	3s	3s	NUM
cana-2550	32	2	(	(	PUNCT
cana-2550	32	3	2025	2025	NUM
cana-2550	32	4	)	)	PUNCT
cana-2550	32	5	69	69	NUM
cana-2550	32	6	https://internationalpubls.com	https://internationalpubls.com	X
cana-2550	32	7	a	a	DET
cana-2550	32	8	multidimensional	multidimensional	ADJ
cana-2550	32	9	,	,	PUNCT
cana-2550	32	10	non	non	ADJ
cana-2550	32	11	-	-	ADJ
cana-2550	32	12	convex	convex	ADJ
cana-2550	32	13	loss	loss	NOUN
cana-2550	32	14	landscape	landscape	NOUN
cana-2550	32	15	with	with	ADP
cana-2550	32	16	multiple	multiple	ADJ
cana-2550	32	17	local	local	ADJ
cana-2550	32	18	minima	minima	NOUN
cana-2550	32	19	,	,	PUNCT
cana-2550	32	20	saddle	saddle	NOUN
cana-2550	32	21	points	point	NOUN
cana-2550	32	22	,	,	PUNCT
cana-2550	32	23	and	and	CCONJ
cana-2550	32	24	plains	plain	NOUN
cana-2550	32	25	.	.	PUNCT
cana-2550	33	1	traditional	traditional	ADJ
cana-2550	33	2	optimization	optimization	NOUN
cana-2550	33	3	techniques	technique	NOUN
cana-2550	33	4	like	like	ADP
cana-2550	33	5	gradient	gradient	ADJ
cana-2550	33	6	descent	descent	NOUN
cana-2550	33	7	(	(	PUNCT
cana-2550	33	8	gd	gd	NOUN
cana-2550	33	9	)	)	PUNCT
cana-2550	33	10	and	and	CCONJ
cana-2550	33	11	its	its	PRON
cana-2550	33	12	variations	variation	NOUN
cana-2550	33	13	are	be	AUX
cana-2550	33	14	pivotal	pivotal	ADJ
cana-2550	33	15	in	in	ADP
cana-2550	33	16	updating	update	VERB
cana-2550	33	17	the	the	DET
cana-2550	33	18	network	network	NOUN
cana-2550	33	19	's	's	PART
cana-2550	33	20	parameters	parameter	NOUN
cana-2550	33	21	iteratively	iteratively	ADV
cana-2550	33	22	to	to	PART
cana-2550	33	23	decrease	decrease	VERB
cana-2550	33	24	the	the	DET
cana-2550	33	25	loss	loss	NOUN
cana-2550	33	26	.	.	PUNCT
cana-2550	34	1	inspite	inspite	ADJ
cana-2550	34	2	of	of	ADP
cana-2550	34	3	this	this	PRON
cana-2550	34	4	they	they	PRON
cana-2550	34	5	also	also	ADV
cana-2550	34	6	have	have	VERB
cana-2550	34	7	multiple	multiple	ADJ
cana-2550	34	8	challenges	challenge	NOUN
cana-2550	34	9	like	like	ADP
cana-2550	34	10	slow	slow	ADJ
cana-2550	34	11	convergence	convergence	NOUN
cana-2550	34	12	,	,	PUNCT
cana-2550	34	13	dependency	dependency	NOUN
cana-2550	34	14	on	on	ADP
cana-2550	34	15	the	the	DET
cana-2550	34	16	hyperparameters	hyperparameter	NOUN
cana-2550	34	17	of	of	ADP
cana-2550	34	18	learning	learn	VERB
cana-2550	34	19	rate	rate	NOUN
cana-2550	34	20	,	,	PUNCT
cana-2550	34	21	and	and	CCONJ
cana-2550	34	22	possibly	possibly	ADV
cana-2550	34	23	being	be	AUX
cana-2550	34	24	trapped	trap	VERB
cana-2550	34	25	in	in	ADP
cana-2550	34	26	suboptimal	suboptimal	ADJ
cana-2550	34	27	solutions	solution	NOUN
cana-2550	34	28	.	.	PUNCT
cana-2550	35	1	mathematics	mathematic	NOUN
cana-2550	35	2	provides	provide	VERB
cana-2550	35	3	a	a	DET
cana-2550	35	4	number	number	NOUN
cana-2550	35	5	of	of	ADP
cana-2550	35	6	solution	solution	NOUN
cana-2550	35	7	to	to	PART
cana-2550	35	8	overcome	overcome	VERB
cana-2550	35	9	these	these	DET
cana-2550	35	10	challenges	challenge	NOUN
cana-2550	35	11	.	.	PUNCT
cana-2550	36	1	variety	variety	NOUN
cana-2550	36	2	of	of	ADP
cana-2550	36	3	important	important	ADJ
cana-2550	36	4	mathematical	mathematical	ADJ
cana-2550	36	5	concepts	concept	NOUN
cana-2550	36	6	of	of	ADP
cana-2550	36	7	linear	linear	PROPN
cana-2550	36	8	algebra	algebra	NOUN
cana-2550	36	9	,	,	PUNCT
cana-2550	36	10	calculus	calculus	NOUN
cana-2550	36	11	,	,	PUNCT
cana-2550	36	12	and	and	CCONJ
cana-2550	36	13	optimization	optimization	NOUN
cana-2550	36	14	theory	theory	NOUN
cana-2550	36	15	are	be	AUX
cana-2550	36	16	necessary	necessary	ADJ
cana-2550	36	17	to	to	PART
cana-2550	36	18	understand	understand	VERB
cana-2550	36	19	and	and	CCONJ
cana-2550	36	20	optimize	optimize	VERB
cana-2550	36	21	the	the	DET
cana-2550	36	22	performance	performance	NOUN
cana-2550	36	23	of	of	ADP
cana-2550	36	24	optimization	optimization	NOUN
cana-2550	36	25	algorithms	algorithm	NOUN
cana-2550	36	26	.	.	PUNCT
cana-2550	37	1	a	a	DET
cana-2550	37	2	number	number	NOUN
cana-2550	37	3	of	of	ADP
cana-2550	37	4	techniques	technique	NOUN
cana-2550	37	5	such	such	ADJ
cana-2550	37	6	as	as	ADP
cana-2550	37	7	stochastic	stochastic	ADJ
cana-2550	37	8	gradient	gradient	ADJ
cana-2550	37	9	descent	descent	NOUN
cana-2550	37	10	,	,	PUNCT
cana-2550	37	11	momentum	momentum	NOUN
cana-2550	37	12	-	-	PUNCT
cana-2550	37	13	based	base	VERB
cana-2550	37	14	methods	method	NOUN
cana-2550	37	15	,	,	PUNCT
cana-2550	37	16	and	and	CCONJ
cana-2550	37	17	adaptive	adaptive	ADJ
cana-2550	37	18	learning	learning	NOUN
cana-2550	37	19	rate	rate	NOUN
cana-2550	37	20	algorithms	algorithm	NOUN
cana-2550	37	21	(	(	PUNCT
cana-2550	37	22	adam	adam	NOUN
cana-2550	37	23	,	,	PUNCT
cana-2550	37	24	rmsprop	rmsprop	NOUN
cana-2550	37	25	)	)	PUNCT
cana-2550	37	26	have	have	AUX
cana-2550	37	27	been	be	AUX
cana-2550	37	28	designed	design	VERB
cana-2550	37	29	specially	specially	ADV
cana-2550	37	30	using	use	VERB
cana-2550	37	31	concepts	concept	NOUN
cana-2550	37	32	of	of	ADP
cana-2550	37	33	mathematics	mathematic	NOUN
cana-2550	37	34	to	to	PART
cana-2550	37	35	improve	improve	VERB
cana-2550	37	36	convergence	convergence	NOUN
cana-2550	37	37	rate	rate	NOUN
cana-2550	37	38	and	and	CCONJ
cana-2550	37	39	stability	stability	NOUN
cana-2550	37	40	.	.	PUNCT
cana-2550	38	1	advanced	advanced	ADJ
cana-2550	38	2	mathematical	mathematical	ADJ
cana-2550	38	3	theory	theory	NOUN
cana-2550	38	4	such	such	ADJ
cana-2550	38	5	as	as	ADP
cana-2550	38	6	hessian	hessian	NOUN
cana-2550	38	7	-	-	PUNCT
cana-2550	38	8	based	base	VERB
cana-2550	38	9	optimization	optimization	NOUN
cana-2550	38	10	and	and	CCONJ
cana-2550	38	11	second	second	ADJ
cana-2550	38	12	-	-	PUNCT
cana-2550	38	13	order	order	NOUN
cana-2550	38	14	techniques	technique	NOUN
cana-2550	38	15	further	far	ADV
cana-2550	38	16	describe	describe	VERB
cana-2550	38	17	curvature	curvature	NOUN
cana-2550	38	18	and	and	CCONJ
cana-2550	38	19	gradient	gradient	NOUN
cana-2550	38	20	to	to	PART
cana-2550	38	21	produce	produce	VERB
cana-2550	38	22	efficient	efficient	ADJ
cana-2550	38	23	optimization	optimization	NOUN
cana-2550	38	24	.	.	PUNCT
cana-2550	39	1	this	this	DET
cana-2550	39	2	paper	paper	NOUN
cana-2550	39	3	helps	help	VERB
cana-2550	39	4	to	to	PART
cana-2550	39	5	understand	understand	VERB
cana-2550	39	6	how	how	SCONJ
cana-2550	39	7	mathematical	mathematical	ADJ
cana-2550	39	8	concepts	concept	NOUN
cana-2550	39	9	are	be	AUX
cana-2550	39	10	used	use	VERB
cana-2550	39	11	to	to	PART
cana-2550	39	12	optimize	optimize	VERB
cana-2550	39	13	the	the	DET
cana-2550	39	14	neural	neural	ADJ
cana-2550	39	15	network	network	NOUN
cana-2550	39	16	,	,	PUNCT
cana-2550	39	17	key	key	ADJ
cana-2550	39	18	techniques	technique	NOUN
cana-2550	39	19	based	base	VERB
cana-2550	39	20	on	on	ADP
cana-2550	39	21	mathematics	mathematic	NOUN
cana-2550	39	22	,	,	PUNCT
cana-2550	39	23	and	and	CCONJ
cana-2550	39	24	approaches	approach	NOUN
cana-2550	39	25	for	for	ADP
cana-2550	39	26	enhancing	enhance	VERB
cana-2550	39	27	deep	deep	ADJ
cana-2550	39	28	learning	learning	NOUN
cana-2550	39	29	architecture	architecture	NOUN
cana-2550	39	30	.	.	PUNCT
cana-2550	40	1	a	a	DET
cana-2550	40	2	deep	deep	ADJ
cana-2550	40	3	analysis	analysis	NOUN
cana-2550	40	4	of	of	ADP
cana-2550	40	5	neural	neural	ADJ
cana-2550	40	6	network	network	NOUN
cana-2550	40	7	optimization	optimization	NOUN
cana-2550	40	8	,	,	PUNCT
cana-2550	40	9	including	include	VERB
cana-2550	40	10	strategies	strategy	NOUN
cana-2550	40	11	related	relate	VERB
cana-2550	40	12	to	to	ADP
cana-2550	40	13	efficiency	efficiency	NOUN
cana-2550	40	14	,	,	PUNCT
cana-2550	40	15	can	can	AUX
cana-2550	40	16	be	be	AUX
cana-2550	40	17	provided	provide	VERB
cana-2550	40	18	based	base	VERB
cana-2550	40	19	on	on	ADP
cana-2550	40	20	mathematical	mathematical	ADJ
cana-2550	40	21	knowledge	knowledge	NOUN
cana-2550	40	22	.	.	PUNCT
cana-2550	41	1	1.1	1.1	NUM
cana-2550	41	2	problem	problem	NOUN
cana-2550	41	3	statement	statement	NOUN
cana-2550	41	4	:	:	PUNCT
cana-2550	41	5	efficient	efficient	ADJ
cana-2550	41	6	training	training	NOUN
cana-2550	41	7	and	and	CCONJ
cana-2550	41	8	optimization	optimization	NOUN
cana-2550	41	9	of	of	ADP
cana-2550	41	10	neural	neural	ADJ
cana-2550	41	11	networks	network	NOUN
cana-2550	41	12	are	be	AUX
cana-2550	41	13	critical	critical	ADJ
cana-2550	41	14	to	to	ADP
cana-2550	41	15	achieving	achieve	VERB
cana-2550	41	16	high	high	ADJ
cana-2550	41	17	performance	performance	NOUN
cana-2550	41	18	without	without	ADP
cana-2550	41	19	incurring	incur	VERB
cana-2550	41	20	excessive	excessive	ADJ
cana-2550	41	21	computational	computational	ADJ
cana-2550	41	22	costs	cost	NOUN
cana-2550	41	23	.	.	PUNCT
cana-2550	42	1	this	this	PRON
cana-2550	42	2	gave	give	VERB
cana-2550	42	3	rise	rise	NOUN
cana-2550	42	4	to	to	ADP
cana-2550	42	5	the	the	DET
cana-2550	42	6	development	development	NOUN
cana-2550	42	7	and	and	CCONJ
cana-2550	42	8	application	application	NOUN
cana-2550	42	9	of	of	ADP
cana-2550	42	10	mathematical	mathematical	ADJ
cana-2550	42	11	optimization	optimization	NOUN
cana-2550	42	12	techniques	technique	NOUN
cana-2550	42	13	tailored	tailor	VERB
cana-2550	42	14	to	to	ADP
cana-2550	42	15	neural	neural	ADJ
cana-2550	42	16	network	network	NOUN
cana-2550	42	17	architectures	architecture	NOUN
cana-2550	42	18	.	.	PUNCT
cana-2550	43	1	1.2	1.2	NUM
cana-2550	43	2	objectives	objective	NOUN
cana-2550	43	3	:	:	PUNCT
cana-2550	43	4	•	•	ADP
cana-2550	43	5	to	to	PART
cana-2550	43	6	explore	explore	VERB
cana-2550	43	7	mathematical	mathematical	ADJ
cana-2550	43	8	foundations	foundation	NOUN
cana-2550	43	9	of	of	ADP
cana-2550	43	10	optimization	optimization	NOUN
cana-2550	43	11	in	in	ADP
cana-2550	43	12	neural	neural	ADJ
cana-2550	43	13	networks	network	NOUN
cana-2550	43	14	.	.	PUNCT
cana-2550	44	1	•	•	ADP
cana-2550	44	2	to	to	PART
cana-2550	44	3	analyze	analyze	VERB
cana-2550	44	4	classical	classical	ADJ
cana-2550	44	5	and	and	CCONJ
cana-2550	44	6	modern	modern	ADJ
cana-2550	44	7	optimization	optimization	NOUN
cana-2550	44	8	techniques	technique	NOUN
cana-2550	44	9	.	.	PUNCT
cana-2550	45	1	•	•	NOUN
cana-2550	45	2	to	to	PART
cana-2550	45	3	propose	propose	VERB
cana-2550	45	4	novel	novel	ADJ
cana-2550	45	5	methods	method	NOUN
cana-2550	45	6	for	for	ADP
cana-2550	45	7	improving	improve	VERB
cana-2550	45	8	efficiency	efficiency	NOUN
cana-2550	45	9	in	in	ADP
cana-2550	45	10	large	large	ADJ
cana-2550	45	11	-	-	PUNCT
cana-2550	45	12	scale	scale	NOUN
cana-2550	45	13	neural	neural	ADJ
cana-2550	45	14	networks	network	NOUN
cana-2550	45	15	.	.	PUNCT
cana-2550	46	1	2	2	X
cana-2550	46	2	.	.	X
cana-2550	46	3	mathematical	mathematical	ADJ
cana-2550	46	4	foundations	foundation	NOUN
cana-2550	46	5	of	of	ADP
cana-2550	46	6	optimization	optimization	NOUN
cana-2550	46	7	in	in	ADP
cana-2550	46	8	neural	neural	ADJ
cana-2550	46	9	networks	network	NOUN
cana-2550	46	10	:	:	PUNCT
cana-2550	46	11	2.1	2.1	NUM
cana-2550	46	12	optimization	optimization	NOUN
cana-2550	46	13	problem	problem	NOUN
cana-2550	46	14	in	in	ADP
cana-2550	46	15	neural	neural	ADJ
cana-2550	46	16	networks	network	NOUN
cana-2550	46	17	:	:	PUNCT
cana-2550	46	18	-the	-the	DET
cana-2550	46	19	main	main	ADJ
cana-2550	46	20	optimization	optimization	NOUN
cana-2550	46	21	challenge	challenge	NOUN
cana-2550	46	22	in	in	ADP
cana-2550	46	23	neural	neural	ADJ
cana-2550	46	24	networks	network	NOUN
cana-2550	46	25	is	be	AUX
cana-2550	46	26	to	to	PART
cana-2550	46	27	determine	determine	VERB
cana-2550	46	28	the	the	DET
cana-2550	46	29	best	good	ADJ
cana-2550	46	30	combination	combination	NOUN
cana-2550	46	31	of	of	ADP
cana-2550	46	32	parameters	parameter	NOUN
cana-2550	46	33	,	,	PUNCT
cana-2550	46	34	including	include	VERB
cana-2550	46	35	weights	weight	NOUN
cana-2550	46	36	and	and	CCONJ
cana-2550	46	37	biases	bias	NOUN
cana-2550	46	38	,	,	PUNCT
cana-2550	46	39	to	to	PART
cana-2550	46	40	minimize	minimize	VERB
cana-2550	46	41	a	a	DET
cana-2550	46	42	specified	specify	VERB
cana-2550	46	43	loss	loss	NOUN
cana-2550	46	44	function	function	NOUN
cana-2550	46	45	.	.	PUNCT
cana-2550	47	1	this	this	DET
cana-2550	47	2	function	function	NOUN
cana-2550	47	3	helps	help	VERB
cana-2550	47	4	to	to	PART
cana-2550	47	5	evaluate	evaluate	VERB
cana-2550	47	6	the	the	DET
cana-2550	47	7	discrepancy	discrepancy	NOUN
cana-2550	47	8	between	between	ADP
cana-2550	47	9	the	the	DET
cana-2550	47	10	predicted	predict	VERB
cana-2550	47	11	result	result	NOUN
cana-2550	47	12	and	and	CCONJ
cana-2550	47	13	the	the	DET
cana-2550	47	14	desired	desire	VERB
cana-2550	47	15	target	target	NOUN
cana-2550	47	16	,	,	PUNCT
cana-2550	47	17	which	which	PRON
cana-2550	47	18	further	far	ADV
cana-2550	47	19	helps	help	VERB
cana-2550	47	20	to	to	PART
cana-2550	47	21	guidew	guidew	VERB
cana-2550	47	22	the	the	DET
cana-2550	47	23	network	network	NOUN
cana-2550	47	24	's	's	PART
cana-2550	47	25	training	training	NOUN
cana-2550	47	26	procedure	procedure	NOUN
cana-2550	47	27	.	.	PUNCT
cana-2550	48	1	the	the	DET
cana-2550	48	2	optimization	optimization	NOUN
cana-2550	48	3	problem	problem	NOUN
cana-2550	48	4	in	in	ADP
cana-2550	48	5	neural	neural	ADJ
cana-2550	48	6	networks	network	NOUN
cana-2550	48	7	is	be	AUX
cana-2550	48	8	typically	typically	ADV
cana-2550	48	9	formulated	formulate	VERB
cana-2550	48	10	as	as	ADP
cana-2550	48	11	:	:	PUNCT
cana-2550	48	12	𝜃∗	𝜃∗	PROPN
cana-2550	48	13	=	=	PUNCT
cana-2550	48	14	𝑎𝑟𝑔	𝑎𝑟𝑔	NOUN
cana-2550	48	15	𝑚𝑖𝑛	𝑚𝑖𝑛	PROPN
cana-2550	48	16	𝜃	𝜃	NOUN
cana-2550	48	17	ℒ	ℒ	X
cana-2550	48	18	(	(	PUNCT
cana-2550	48	19	𝜃	𝜃	NOUN
cana-2550	48	20	;	;	PUNCT
cana-2550	48	21	𝑋	𝑋	PROPN
cana-2550	48	22	,	,	PUNCT
cana-2550	48	23	𝑌	𝑌	PROPN
cana-2550	48	24	)	)	PUNCT
cana-2550	48	25	where	where	SCONJ
cana-2550	48	26	𝐿	𝐿	PROPN
cana-2550	48	27	𝑖s	𝑖s	ADP
cana-2550	48	28	the	the	DET
cana-2550	48	29	loss	loss	NOUN
cana-2550	48	30	function	function	NOUN
cana-2550	48	31	,	,	PUNCT
cana-2550	48	32	θ	θ	PROPN
cana-2550	48	33	represents	represent	VERB
cana-2550	48	34	the	the	DET
cana-2550	48	35	model	model	NOUN
cana-2550	48	36	parameters	parameter	NOUN
cana-2550	48	37	,	,	PUNCT
cana-2550	48	38	and	and	CCONJ
cana-2550	48	39	(	(	PUNCT
cana-2550	48	40	x	x	X
cana-2550	48	41	,	,	PUNCT
cana-2550	48	42	y	y	NOUN
cana-2550	48	43	)	)	PUNCT
cana-2550	48	44	are	be	AUX
cana-2550	48	45	the	the	DET
cana-2550	48	46	input	input	NOUN
cana-2550	48	47	-	-	PUNCT
cana-2550	48	48	output	output	NOUN
cana-2550	48	49	pairs	pair	NOUN
cana-2550	48	50	.	.	PUNCT
cana-2550	49	1	2.2	2.2	NUM
cana-2550	49	2	loss	loss	NOUN
cana-2550	49	3	functions	function	NOUN
cana-2550	49	4	:	:	PUNCT
cana-2550	49	5	the	the	DET
cana-2550	49	6	loss	loss	NOUN
cana-2550	49	7	function	function	NOUN
cana-2550	49	8	varies	vary	VERB
cana-2550	49	9	based	base	VERB
cana-2550	49	10	on	on	ADP
cana-2550	49	11	the	the	DET
cana-2550	49	12	task	task	NOUN
cana-2550	49	13	;	;	PUNCT
cana-2550	49	14	for	for	ADP
cana-2550	49	15	instance	instance	NOUN
cana-2550	49	16	,	,	PUNCT
cana-2550	49	17	the	the	DET
cana-2550	49	18	mean	mean	NOUN
cana-2550	49	19	squared	square	VERB
cana-2550	49	20	error	error	NOUN
cana-2550	49	21	(	(	PUNCT
cana-2550	49	22	mse	mse	NOUN
cana-2550	49	23	)	)	PUNCT
cana-2550	49	24	is	be	AUX
cana-2550	49	25	used	use	VERB
cana-2550	49	26	in	in	ADP
cana-2550	49	27	regression	regression	NOUN
cana-2550	49	28	tasks	task	NOUN
cana-2550	49	29	,	,	PUNCT
cana-2550	49	30	while	while	SCONJ
cana-2550	49	31	cross	cross	ADJ
cana-2550	49	32	-	-	ADJ
cana-2550	49	33	entropy	entropy	ADJ
cana-2550	49	34	loss	loss	NOUN
cana-2550	49	35	is	be	AUX
cana-2550	49	36	common	common	ADJ
cana-2550	49	37	in	in	ADP
cana-2550	49	38	classification	classification	NOUN
cana-2550	49	39	tasks	task	NOUN
cana-2550	49	40	.	.	PUNCT
cana-2550	50	1	2.2.a	2.2.a	NUM
cana-2550	50	2	mean	mean	VERB
cana-2550	50	3	squared	square	VERB
cana-2550	50	4	error	error	NOUN
cana-2550	50	5	(	(	PUNCT
cana-2550	50	6	mse	mse	NOUN
cana-2550	50	7	):	):	PUNCT
cana-2550	50	8	the	the	DET
cana-2550	50	9	mean	mean	ADJ
cana-2550	50	10	squared	square	VERB
cana-2550	50	11	error	error	NOUN
cana-2550	50	12	(	(	PUNCT
cana-2550	50	13	mse	mse	NOUN
cana-2550	50	14	)	)	PUNCT
cana-2550	50	15	serves	serve	VERB
cana-2550	50	16	as	as	ADP
cana-2550	50	17	a	a	DET
cana-2550	50	18	widely	widely	ADV
cana-2550	50	19	employed	employ	VERB
cana-2550	50	20	loss	loss	NOUN
cana-2550	50	21	function	function	NOUN
cana-2550	50	22	in	in	ADP
cana-2550	50	23	regression	regression	NOUN
cana-2550	50	24	assignments	assignment	NOUN
cana-2550	50	25	,	,	PUNCT
cana-2550	50	26	which	which	PRON
cana-2550	50	27	computes	compute	VERB
cana-2550	50	28	the	the	DET
cana-2550	50	29	mean	mean	ADJ
cana-2550	50	30	squared	square	VERB
cana-2550	50	31	variance	variance	NOUN
cana-2550	50	32	between	between	ADP
cana-2550	50	33	the	the	DET
cana-2550	50	34	predicted	predict	VERB
cana-2550	50	35	values	value	NOUN
cana-2550	50	36	and	and	CCONJ
cana-2550	50	37	the	the	DET
cana-2550	50	38	actual	actual	ADJ
cana-2550	50	39	values	value	NOUN
cana-2550	50	40	.	.	PUNCT
cana-2550	51	1	this	this	PRON
cana-2550	51	2	is	be	AUX
cana-2550	51	3	required	require	VERB
cana-2550	51	4	to	to	PART
cana-2550	51	5	calcuate	calcuate	VERB
cana-2550	51	6	the	the	DET
cana-2550	51	7	average	average	NOUN
cana-2550	51	8	of	of	ADP
cana-2550	51	9	the	the	DET
cana-2550	51	10	squared	squared	ADJ
cana-2550	51	11	variations	variation	NOUN
cana-2550	51	12	across	across	ADP
cana-2550	51	13	all	all	DET
cana-2550	51	14	data	datum	NOUN
cana-2550	51	15	points	point	NOUN
cana-2550	51	16	.	.	PUNCT
cana-2550	52	1	mse	mse	PROPN
cana-2550	52	2	is	be	AUX
cana-2550	52	3	particularly	particularly	ADV
cana-2550	52	4	responsive	responsive	ADJ
cana-2550	52	5	to	to	ADP
cana-2550	52	6	significant	significant	ADJ
cana-2550	52	7	errors	error	NOUN
cana-2550	52	8	because	because	SCONJ
cana-2550	52	9	of	of	ADP
cana-2550	52	10	the	the	DET
cana-2550	52	11	squaring	square	VERB
cana-2550	52	12	process	process	NOUN
cana-2550	52	13	,	,	PUNCT
cana-2550	52	14	thus	thus	ADV
cana-2550	52	15	effectively	effectively	ADV
cana-2550	52	16	correcting	correct	VERB
cana-2550	52	17	substantial	substantial	ADJ
cana-2550	52	18	deviations	deviation	NOUN
cana-2550	52	19	from	from	ADP
cana-2550	52	20	the	the	DET
cana-2550	52	21	true	true	ADJ
cana-2550	52	22	values	value	NOUN
cana-2550	52	23	.	.	PUNCT
cana-2550	53	1	communications	communication	NOUN
cana-2550	53	2	on	on	ADP
cana-2550	53	3	applied	apply	VERB
cana-2550	53	4	nonlinear	nonlinear	ADJ
cana-2550	53	5	analysis	analysis	NOUN
cana-2550	53	6	issn	issn	NOUN
cana-2550	53	7	:	:	PUNCT
cana-2550	53	8	1074	1074	NUM
cana-2550	53	9	-	-	PUNCT
cana-2550	53	10	133x	133x	NUM
cana-2550	53	11	vol	vol	NOUN
cana-2550	53	12	32	32	NUM
cana-2550	53	13	no	no	NOUN
cana-2550	53	14	.	.	PUNCT
cana-2550	54	1	3s	3s	NUM
cana-2550	54	2	(	(	PUNCT
cana-2550	54	3	2025	2025	NUM
cana-2550	54	4	)	)	PUNCT
cana-2550	54	5	70	70	NUM
cana-2550	54	6	https://internationalpubls.com	https://internationalpubls.com	X
cana-2550	54	7	ℒ	ℒ	NOUN
cana-2550	54	8	=	=	SYM
cana-2550	54	9	1	1	NUM
cana-2550	54	10	𝑁	𝑁	NOUN
cana-2550	54	11	∑(𝑦𝑖	∑(𝑦𝑖	PROPN
cana-2550	54	12	−	−	PROPN
cana-2550	54	13	𝑦	𝑦	SYM
cana-2550	54	14	�	�	PROPN
cana-2550	54	15	̂	̂	SYM
cana-2550	54	16	�	�	NOUN
cana-2550	54	17	)	)	PUNCT
cana-2550	54	18	2	2	NUM
cana-2550	55	1	𝑁	𝑁	PROPN
cana-2550	55	2	𝑖=1	𝑖=1	PROPN
cana-2550	55	3	n	n	X
cana-2550	55	4	is	be	AUX
cana-2550	55	5	the	the	DET
cana-2550	55	6	number	number	NOUN
cana-2550	55	7	of	of	ADP
cana-2550	55	8	data	datum	NOUN
cana-2550	55	9	points	point	NOUN
cana-2550	55	10	𝑦𝑖	𝑦𝑖	PROPN
cana-2550	55	11	is	be	AUX
cana-2550	55	12	the	the	DET
cana-2550	55	13	true	true	ADJ
cana-2550	55	14	value	value	NOUN
cana-2550	55	15	.	.	PUNCT
cana-2550	56	1	𝑦	𝑦	X
cana-2550	56	2	�	�	PROPN
cana-2550	56	3	̂	̂	VERB
cana-2550	56	4	�	�	NOUN
cana-2550	56	5	is	be	AUX
cana-2550	56	6	the	the	DET
cana-2550	56	7	predicted	predict	VERB
cana-2550	56	8	value	value	NOUN
cana-2550	56	9	.	.	PUNCT
cana-2550	57	1	2.2.b	2.2.b	NUM
cana-2550	57	2	cross	cross	NOUN
cana-2550	57	3	entropy	entropy	PROPN
cana-2550	57	4	loss	loss	PROPN
cana-2550	57	5	:	:	PUNCT
cana-2550	57	6	cross	cross	ADJ
cana-2550	57	7	-	-	ADJ
cana-2550	57	8	entropy	entropy	NOUN
cana-2550	57	9	serves	serve	VERB
cana-2550	57	10	as	as	ADP
cana-2550	57	11	a	a	DET
cana-2550	57	12	common	common	ADJ
cana-2550	57	13	loss	loss	NOUN
cana-2550	57	14	function	function	NOUN
cana-2550	57	15	in	in	ADP
cana-2550	57	16	classification	classification	NOUN
cana-2550	57	17	assignments	assignment	NOUN
cana-2550	57	18	,	,	PUNCT
cana-2550	57	19	which	which	PRON
cana-2550	57	20	helps	help	VERB
cana-2550	57	21	to	to	PART
cana-2550	57	22	compute	compute	VERB
cana-2550	57	23	the	the	DET
cana-2550	57	24	variance	variance	NOUN
cana-2550	57	25	between	between	ADP
cana-2550	57	26	the	the	DET
cana-2550	57	27	actual	actual	ADJ
cana-2550	57	28	label	label	NOUN
cana-2550	57	29	distribution	distribution	NOUN
cana-2550	57	30	and	and	CCONJ
cana-2550	57	31	the	the	DET
cana-2550	57	32	proposed	propose	VERB
cana-2550	57	33	possibility	possibility	NOUN
cana-2550	57	34	distribution	distribution	NOUN
cana-2550	57	35	.	.	PUNCT
cana-2550	58	1	it	it	PRON
cana-2550	58	2	corrects	correct	VERB
cana-2550	58	3	inaccuracies	inaccuracy	NOUN
cana-2550	58	4	by	by	ADP
cana-2550	58	5	moving	move	VERB
cana-2550	58	6	the	the	DET
cana-2550	58	7	loss	loss	NOUN
cana-2550	58	8	when	when	SCONJ
cana-2550	58	9	the	the	DET
cana-2550	58	10	predicted	predict	VERB
cana-2550	58	11	probabilities	probability	NOUN
cana-2550	58	12	deviate	deviate	VERB
cana-2550	58	13	from	from	ADP
cana-2550	58	14	the	the	DET
cana-2550	58	15	true	true	ADJ
cana-2550	58	16	labels	label	NOUN
cana-2550	58	17	.	.	PUNCT
cana-2550	59	1	in	in	ADP
cana-2550	59	2	mathematical	mathematical	ADJ
cana-2550	59	3	terms	term	NOUN
cana-2550	59	4	,	,	PUNCT
cana-2550	59	5	it	it	PRON
cana-2550	59	6	represents	represent	VERB
cana-2550	59	7	the	the	DET
cana-2550	59	8	negative	negative	ADJ
cana-2550	59	9	loglikelihood	loglikelihood	NOUN
cana-2550	59	10	of	of	ADP
cana-2550	59	11	the	the	DET
cana-2550	59	12	accurate	accurate	ADJ
cana-2550	59	13	category	category	NOUN
cana-2550	59	14	,	,	PUNCT
cana-2550	59	15	computed	compute	VERB
cana-2550	59	16	across	across	ADP
cana-2550	59	17	all	all	DET
cana-2550	59	18	data	datum	NOUN
cana-2550	59	19	points	point	NOUN
cana-2550	59	20	.	.	PUNCT
cana-2550	60	1	ℒcosnrp	ℒcosnrp	PROPN
cana-2550	60	2	=	=	PUNCT
cana-2550	61	1	−	−	PROPN
cana-2550	61	2	1	1	NUM
cana-2550	61	3	n	n	PROPN
cana-2550	61	4	∑	∑	ADP
cana-2550	61	5	yi	yi	PROPN
cana-2550	61	6	n	n	PROPN
cana-2550	61	7	i=1	i=1	PROPN
cana-2550	61	8	log(yî	log(yî	PROPN
cana-2550	61	9	)	)	PUNCT
cana-2550	62	1	n	n	CCONJ
cana-2550	62	2	is	be	AUX
cana-2550	62	3	the	the	DET
cana-2550	62	4	number	number	NOUN
cana-2550	62	5	of	of	ADP
cana-2550	62	6	data	datum	NOUN
cana-2550	62	7	points	point	NOUN
cana-2550	62	8	,	,	PUNCT
cana-2550	62	9	yi	yi	PROPN
cana-2550	62	10	is	be	AUX
cana-2550	62	11	the	the	DET
cana-2550	62	12	true	true	ADJ
cana-2550	62	13	probability	probability	NOUN
cana-2550	62	14	yî	yî	PROPN
cana-2550	62	15	is	be	AUX
cana-2550	62	16	the	the	DET
cana-2550	62	17	predicted	predict	VERB
cana-2550	62	18	probability	probability	NOUN
cana-2550	62	19	.	.	PUNCT
cana-2550	63	1	2.2.c	2.2.c	NUM
cana-2550	63	2	gradient	gradient	NOUN
cana-2550	63	3	based	base	VERB
cana-2550	63	4	optimization	optimization	NOUN
cana-2550	63	5	:	:	PUNCT
cana-2550	63	6	gradient	gradient	NOUN
cana-2550	63	7	-	-	PUNCT
cana-2550	63	8	based	base	VERB
cana-2550	63	9	optimization	optimization	NOUN
cana-2550	63	10	is	be	AUX
cana-2550	63	11	a	a	DET
cana-2550	63	12	vital	vital	ADJ
cana-2550	63	13	technique	technique	NOUN
cana-2550	63	14	,	,	PUNCT
cana-2550	63	15	causing	cause	VERB
cana-2550	63	16	iteratively	iteratively	ADV
cana-2550	63	17	fine	fine	ADJ
cana-2550	63	18	-	-	PUNCT
cana-2550	63	19	tuning	tune	VERB
cana-2550	63	20	model	model	NOUN
cana-2550	63	21	parameters	parameter	NOUN
cana-2550	63	22	by	by	ADP
cana-2550	63	23	moving	move	VERB
cana-2550	63	24	along	along	ADP
cana-2550	63	25	the	the	DET
cana-2550	63	26	opposite	opposite	ADJ
cana-2550	63	27	direction	direction	NOUN
cana-2550	63	28	of	of	ADP
cana-2550	63	29	the	the	DET
cana-2550	63	30	negative	negative	ADJ
cana-2550	63	31	gradient	gradient	NOUN
cana-2550	63	32	of	of	ADP
cana-2550	63	33	the	the	DET
cana-2550	63	34	loss	loss	NOUN
cana-2550	63	35	function	function	NOUN
cana-2550	63	36	in	in	ADP
cana-2550	63	37	order	order	NOUN
cana-2550	63	38	to	to	PART
cana-2550	63	39	reduce	reduce	VERB
cana-2550	63	40	it	it	PRON
cana-2550	63	41	.	.	PUNCT
cana-2550	64	1	the	the	DET
cana-2550	64	2	gradient	gradient	NOUN
cana-2550	64	3	helps	help	VERB
cana-2550	64	4	to	to	PART
cana-2550	64	5	identify	identify	VERB
cana-2550	64	6	how	how	SCONJ
cana-2550	64	7	quickly	quickly	ADV
cana-2550	64	8	loss	loss	VERB
cana-2550	64	9	changes	change	NOUN
cana-2550	64	10	with	with	ADP
cana-2550	64	11	respect	respect	NOUN
cana-2550	64	12	to	to	ADP
cana-2550	64	13	parameters	parameter	NOUN
cana-2550	64	14	,	,	PUNCT
cana-2550	64	15	which	which	PRON
cana-2550	64	16	promotes	promote	VERB
cana-2550	64	17	guiding	guide	VERB
cana-2550	64	18	the	the	DET
cana-2550	64	19	optimization	optimization	NOUN
cana-2550	64	20	process	process	NOUN
cana-2550	64	21	.	.	PUNCT
cana-2550	65	1	various	various	ADJ
cana-2550	65	2	algorithms	algorithm	NOUN
cana-2550	65	3	for	for	ADP
cana-2550	65	4	example	example	NOUN
cana-2550	65	5	gradient	gradient	NOUN
cana-2550	65	6	descent	descent	NOUN
cana-2550	65	7	which	which	PRON
cana-2550	65	8	is	be	AUX
cana-2550	65	9	commonly	commonly	ADV
cana-2550	65	10	used	use	VERB
cana-2550	65	11	across	across	ADP
cana-2550	65	12	the	the	DET
cana-2550	65	13	board	board	NOUN
cana-2550	65	14	,	,	PUNCT
cana-2550	65	15	depend	depend	VERB
cana-2550	65	16	upon	upon	SCONJ
cana-2550	65	17	this	this	DET
cana-2550	65	18	method	method	NOUN
cana-2550	65	19	to	to	PART
cana-2550	65	20	resourcefully	resourcefully	ADV
cana-2550	65	21	improve	improve	VERB
cana-2550	65	22	the	the	DET
cana-2550	65	23	performance	performance	NOUN
cana-2550	65	24	of	of	ADP
cana-2550	65	25	neural	neural	ADJ
cana-2550	65	26	networks	network	NOUN
cana-2550	65	27	and	and	CCONJ
cana-2550	65	28	other	other	ADJ
cana-2550	65	29	machine	machine	NOUN
cana-2550	65	30	learning	learning	NOUN
cana-2550	65	31	models	model	NOUN
cana-2550	65	32	.	.	PUNCT
cana-2550	66	1	θ𝑡+1	θ𝑡+1	NOUN
cana-2550	66	2	=	=	SYM
cana-2550	66	3	θ𝑡	θ𝑡	PROPN
cana-2550	66	4	−	−	PROPN
cana-2550	66	5	η∇θℒ(θ𝑡	η∇θℒ(θ𝑡	PROPN
cana-2550	66	6	)	)	PUNCT
cana-2550	67	1	𝜃𝑡	𝜃𝑡	VERB
cana-2550	67	2	represents	represent	VERB
cana-2550	67	3	the	the	DET
cana-2550	67	4	parameters	parameter	NOUN
cana-2550	67	5	at	at	ADP
cana-2550	67	6	iteration	iteration	PROPN
cana-2550	67	7	t	t	PROPN
cana-2550	67	8	,	,	PUNCT
cana-2550	67	9	𝜂	𝜂	PROPN
cana-2550	67	10	is	be	AUX
cana-2550	67	11	the	the	DET
cana-2550	67	12	learning	learning	NOUN
cana-2550	67	13	rate	rate	NOUN
cana-2550	67	14	,	,	PUNCT
cana-2550	67	15	∇𝜃ℒ(𝜃𝑡	∇𝜃ℒ(𝜃𝑡	NOUN
cana-2550	67	16	)	)	PUNCT
cana-2550	67	17	is	be	AUX
cana-2550	67	18	the	the	DET
cana-2550	67	19	gradient	gradient	NOUN
cana-2550	67	20	of	of	ADP
cana-2550	67	21	the	the	DET
cana-2550	67	22	loss	loss	NOUN
cana-2550	67	23	function	function	NOUN
cana-2550	67	24	with	with	ADP
cana-2550	67	25	respect	respect	NOUN
cana-2550	67	26	to	to	ADP
cana-2550	67	27	the	the	DET
cana-2550	67	28	parameters	parameter	NOUN
cana-2550	67	29	at	at	ADP
cana-2550	67	30	iteration	iteration	NOUN
cana-2550	67	31	t.	t.	PROPN
cana-2550	67	32	3	3	NUM
cana-2550	67	33	.	.	PUNCT
cana-2550	67	34	classical	classical	ADJ
cana-2550	67	35	optimization	optimization	NOUN
cana-2550	67	36	technique	technique	NOUN
cana-2550	67	37	:	:	PUNCT
cana-2550	67	38	these	these	DET
cana-2550	67	39	techniques	technique	NOUN
cana-2550	67	40	are	be	AUX
cana-2550	67	41	the	the	DET
cana-2550	67	42	traditional	traditional	ADJ
cana-2550	67	43	methods	method	NOUN
cana-2550	67	44	used	use	VERB
cana-2550	67	45	for	for	ADP
cana-2550	67	46	finding	find	VERB
cana-2550	67	47	the	the	DET
cana-2550	67	48	best	good	ADJ
cana-2550	67	49	solution	solution	NOUN
cana-2550	67	50	of	of	ADP
cana-2550	67	51	a	a	DET
cana-2550	67	52	problem	problem	NOUN
cana-2550	67	53	,	,	PUNCT
cana-2550	67	54	which	which	PRON
cana-2550	67	55	in	in	ADP
cana-2550	67	56	most	most	ADJ
cana-2550	67	57	cases	case	NOUN
cana-2550	67	58	is	be	AUX
cana-2550	67	59	achieved	achieve	VERB
cana-2550	67	60	either	either	CCONJ
cana-2550	67	61	by	by	ADP
cana-2550	67	62	minimizing	minimize	VERB
cana-2550	67	63	or	or	CCONJ
cana-2550	67	64	maximizing	maximize	VERB
cana-2550	67	65	some	some	DET
cana-2550	67	66	specified	specify	VERB
cana-2550	67	67	objective	objective	ADJ
cana-2550	67	68	function	function	NOUN
cana-2550	67	69	.	.	PUNCT
cana-2550	68	1	some	some	PRON
cana-2550	68	2	of	of	ADP
cana-2550	68	3	the	the	DET
cana-2550	68	4	traditional	traditional	ADJ
cana-2550	68	5	optimization	optimization	NOUN
cana-2550	68	6	techniques	technique	NOUN
cana-2550	68	7	are	be	AUX
cana-2550	68	8	gradient	gradient	ADJ
cana-2550	68	9	descent	descent	NOUN
cana-2550	68	10	,	,	PUNCT
cana-2550	68	11	where	where	SCONJ
cana-2550	68	12	the	the	DET
cana-2550	68	13	parameters	parameter	NOUN
cana-2550	68	14	are	be	AUX
cana-2550	68	15	changed	change	VERB
cana-2550	68	16	according	accord	VERB
cana-2550	68	17	to	to	ADP
cana-2550	68	18	the	the	DET
cana-2550	68	19	gradient	gradient	NOUN
cana-2550	68	20	of	of	ADP
cana-2550	68	21	the	the	DET
cana-2550	68	22	function	function	NOUN
cana-2550	68	23	,	,	PUNCT
cana-2550	68	24	and	and	CCONJ
cana-2550	68	25	newton	newton	PROPN
cana-2550	68	26	's	's	PART
cana-2550	68	27	method	method	NOUN
cana-2550	68	28	,	,	PUNCT
cana-2550	68	29	which	which	PRON
cana-2550	68	30	uses	use	VERB
cana-2550	68	31	the	the	DET
cana-2550	68	32	second	second	ADJ
cana-2550	68	33	-	-	PUNCT
cana-2550	68	34	order	order	NOUN
cana-2550	68	35	derivatives	derivative	NOUN
cana-2550	68	36	to	to	PART
cana-2550	68	37	speed	speed	VERB
cana-2550	68	38	up	up	ADP
cana-2550	68	39	convergence	convergence	NOUN
cana-2550	68	40	.	.	PUNCT
cana-2550	69	1	other	other	ADJ
cana-2550	69	2	methods	method	NOUN
cana-2550	69	3	like	like	ADP
cana-2550	69	4	conjugate	conjugate	ADJ
cana-2550	69	5	gradient	gradient	NOUN
cana-2550	69	6	and	and	CCONJ
cana-2550	69	7	quasi	quasi	ADJ
cana-2550	69	8	-	-	ADJ
cana-2550	69	9	newton	newton	PROPN
cana-2550	69	10	methods	method	NOUN
cana-2550	69	11	helps	help	VERB
cana-2550	69	12	to	to	PART
cana-2550	69	13	balance	balance	VERB
cana-2550	69	14	computational	computational	ADJ
cana-2550	69	15	efficiency	efficiency	NOUN
cana-2550	69	16	with	with	ADP
cana-2550	69	17	accuracy	accuracy	NOUN
cana-2550	69	18	and	and	CCONJ
cana-2550	69	19	proves	prove	VERB
cana-2550	69	20	to	to	PART
cana-2550	69	21	be	be	AUX
cana-2550	69	22	beneficial	beneficial	ADJ
cana-2550	69	23	in	in	ADP
cana-2550	69	24	tackling	tackle	VERB
cana-2550	69	25	high	high	ADJ
cana-2550	69	26	-	-	PUNCT
cana-2550	69	27	dimensional	dimensional	ADJ
cana-2550	69	28	optimization	optimization	NOUN
cana-2550	69	29	problems	problem	NOUN
cana-2550	69	30	.	.	PUNCT
cana-2550	70	1	beyond	beyond	ADP
cana-2550	70	2	their	their	PRON
cana-2550	70	3	many	many	ADJ
cana-2550	70	4	benefits	benefit	NOUN
cana-2550	70	5	,	,	PUNCT
cana-2550	70	6	these	these	DET
cana-2550	70	7	methods	method	NOUN
cana-2550	70	8	often	often	ADV
cana-2550	70	9	find	find	VERB
cana-2550	70	10	it	it	PRON
cana-2550	70	11	challenging	challenge	VERB
cana-2550	70	12	to	to	PART
cana-2550	70	13	handle	handle	VERB
cana-2550	70	14	huge	huge	ADJ
cana-2550	70	15	neural	neural	ADJ
cana-2550	70	16	networks	network	NOUN
cana-2550	70	17	due	due	ADP
cana-2550	70	18	to	to	ADP
cana-2550	70	19	nonconvex	nonconvex	NOUN
cana-2550	70	20	structures	structure	NOUN
cana-2550	70	21	in	in	ADP
cana-2550	70	22	the	the	DET
cana-2550	70	23	problem	problem	NOUN
cana-2550	70	24	and	and	CCONJ
cana-2550	70	25	complex	complex	ADJ
cana-2550	70	26	parameter	parameter	NOUN
cana-2550	70	27	space	space	NOUN
cana-2550	70	28	.	.	PUNCT
cana-2550	71	1	communications	communication	NOUN
cana-2550	71	2	on	on	ADP
cana-2550	71	3	applied	apply	VERB
cana-2550	71	4	nonlinear	nonlinear	ADJ
cana-2550	71	5	analysis	analysis	NOUN
cana-2550	71	6	issn	issn	NOUN
cana-2550	71	7	:	:	PUNCT
cana-2550	71	8	1074	1074	NUM
cana-2550	71	9	-	-	PUNCT
cana-2550	71	10	133x	133x	NUM
cana-2550	71	11	vol	vol	NOUN
cana-2550	71	12	32	32	NUM
cana-2550	71	13	no	no	NOUN
cana-2550	71	14	.	.	PUNCT
cana-2550	72	1	3s	3s	NUM
cana-2550	72	2	(	(	PUNCT
cana-2550	72	3	2025	2025	NUM
cana-2550	72	4	)	)	PUNCT
cana-2550	72	5	71	71	NUM
cana-2550	72	6	https://internationalpubls.com	https://internationalpubls.com	SYM
cana-2550	72	7	3.1	3.1	NUM
cana-2550	72	8	gradient	gradient	ADJ
cana-2550	72	9	descent	descent	NOUN
cana-2550	72	10	variants	variant	NOUN
cana-2550	72	11	:	:	PUNCT
cana-2550	72	12	this	this	PRON
cana-2550	72	13	is	be	AUX
cana-2550	72	14	an	an	DET
cana-2550	72	15	iterative	iterative	ADJ
cana-2550	72	16	optimization	optimization	NOUN
cana-2550	72	17	procedure	procedure	NOUN
cana-2550	72	18	which	which	PRON
cana-2550	72	19	is	be	AUX
cana-2550	72	20	used	use	VERB
cana-2550	72	21	to	to	PART
cana-2550	72	22	reduce	reduce	VERB
cana-2550	72	23	a	a	DET
cana-2550	72	24	loss	loss	NOUN
cana-2550	72	25	function	function	NOUN
cana-2550	72	26	by	by	ADP
cana-2550	72	27	tuning	tune	VERB
cana-2550	72	28	the	the	DET
cana-2550	72	29	model	model	NOUN
cana-2550	72	30	's	's	PART
cana-2550	72	31	parameters	parameter	NOUN
cana-2550	72	32	towards	towards	ADP
cana-2550	72	33	the	the	DET
cana-2550	72	34	negative	negative	ADJ
cana-2550	72	35	gradient	gradient	NOUN
cana-2550	72	36	's	's	PART
cana-2550	72	37	direction	direction	NOUN
cana-2550	72	38	.	.	PUNCT
cana-2550	73	1	the	the	DET
cana-2550	73	2	gradient	gradient	NOUN
cana-2550	73	3	represents	represent	VERB
cana-2550	73	4	how	how	SCONJ
cana-2550	73	5	the	the	DET
cana-2550	73	6	loss	loss	NOUN
cana-2550	73	7	function	function	NOUN
cana-2550	73	8	changes	change	VERB
cana-2550	73	9	about	about	ADP
cana-2550	73	10	each	each	DET
cana-2550	73	11	parameter	parameter	NOUN
cana-2550	73	12	.	.	PUNCT
cana-2550	74	1	random	random	ADJ
cana-2550	74	2	parameters	parameter	NOUN
cana-2550	74	3	are	be	AUX
cana-2550	74	4	given	give	VERB
cana-2550	74	5	to	to	PART
cana-2550	74	6	start	start	VERB
cana-2550	74	7	the	the	DET
cana-2550	74	8	algorithm	algorithm	NOUN
cana-2550	74	9	and	and	CCONJ
cana-2550	74	10	then	then	ADV
cana-2550	74	11	it	it	PRON
cana-2550	74	12	is	be	AUX
cana-2550	74	13	updated	update	VERB
cana-2550	74	14	based	base	VERB
cana-2550	74	15	on	on	ADP
cana-2550	74	16	the	the	DET
cana-2550	74	17	gradient	gradient	NOUN
cana-2550	74	18	computed	compute	VERB
cana-2550	74	19	at	at	ADP
cana-2550	74	20	each	each	DET
cana-2550	74	21	iteration	iteration	NOUN
cana-2550	74	22	.	.	PUNCT
cana-2550	75	1	the	the	DET
cana-2550	75	2	benefit	benefit	NOUN
cana-2550	75	3	of	of	ADP
cana-2550	75	4	using	use	VERB
cana-2550	75	5	this	this	DET
cana-2550	75	6	algorithm	algorithm	NOUN
cana-2550	75	7	is	be	AUX
cana-2550	75	8	that	that	SCONJ
cana-2550	75	9	it	it	PRON
cana-2550	75	10	is	be	AUX
cana-2550	75	11	very	very	ADV
cana-2550	75	12	easy	easy	ADJ
cana-2550	75	13	to	to	PART
cana-2550	75	14	use	use	VERB
cana-2550	75	15	and	and	CCONJ
cana-2550	75	16	implement	implement	VERB
cana-2550	75	17	.	.	PUNCT
cana-2550	76	1	however	however	ADV
cana-2550	76	2	,	,	PUNCT
cana-2550	76	3	this	this	DET
cana-2550	76	4	method	method	NOUN
cana-2550	76	5	can	can	AUX
cana-2550	76	6	be	be	AUX
cana-2550	76	7	expensive	expensive	ADJ
cana-2550	76	8	in	in	ADP
cana-2550	76	9	terms	term	NOUN
cana-2550	76	10	of	of	ADP
cana-2550	76	11	calculations	calculation	NOUN
cana-2550	76	12	for	for	ADP
cana-2550	76	13	large	large	ADJ
cana-2550	76	14	datasets	dataset	NOUN
cana-2550	76	15	,	,	PUNCT
cana-2550	76	16	as	as	SCONJ
cana-2550	76	17	it	it	PRON
cana-2550	76	18	requires	require	VERB
cana-2550	76	19	the	the	DET
cana-2550	76	20	entire	entire	ADJ
cana-2550	76	21	dataset	dataset	NOUN
cana-2550	76	22	to	to	PART
cana-2550	76	23	compute	compute	VERB
cana-2550	76	24	gradients	gradient	NOUN
cana-2550	76	25	at	at	ADP
cana-2550	76	26	each	each	DET
cana-2550	76	27	step	step	NOUN
cana-2550	76	28	.	.	PUNCT
cana-2550	77	1	to	to	PART
cana-2550	77	2	overcome	overcome	VERB
cana-2550	77	3	this	this	DET
cana-2550	77	4	challenge	challenge	NOUN
cana-2550	77	5	,	,	PUNCT
cana-2550	77	6	several	several	ADJ
cana-2550	77	7	variants	variant	NOUN
cana-2550	77	8	of	of	ADP
cana-2550	77	9	standard	standard	ADJ
cana-2550	77	10	gradient	gradient	ADJ
cana-2550	77	11	descent	descent	NOUN
cana-2550	77	12	are	be	AUX
cana-2550	77	13	developed	develop	VERB
cana-2550	77	14	.	.	PUNCT
cana-2550	78	1	3.1.a	3.1.a	NUM
cana-2550	78	2	stochastic	stochastic	ADJ
cana-2550	78	3	gradient	gradient	ADJ
cana-2550	78	4	descent	descent	NOUN
cana-2550	78	5	(	(	PUNCT
cana-2550	78	6	sgd	sgd	PROPN
cana-2550	78	7	):	):	PUNCT
cana-2550	78	8	this	this	DET
cana-2550	78	9	method	method	NOUN
cana-2550	78	10	is	be	AUX
cana-2550	78	11	derived	derive	VERB
cana-2550	78	12	by	by	ADP
cana-2550	78	13	making	make	VERB
cana-2550	78	14	some	some	DET
cana-2550	78	15	modification	modification	NOUN
cana-2550	78	16	of	of	ADP
cana-2550	78	17	gradient	gradient	ADJ
cana-2550	78	18	descent	descent	NOUN
cana-2550	78	19	,	,	PUNCT
cana-2550	78	20	adjusting	adjust	VERB
cana-2550	78	21	model	model	NOUN
cana-2550	78	22	parameters	parameter	NOUN
cana-2550	78	23	after	after	ADP
cana-2550	78	24	handling	handle	VERB
cana-2550	78	25	each	each	DET
cana-2550	78	26	training	training	NOUN
cana-2550	78	27	sample	sample	NOUN
cana-2550	78	28	,	,	PUNCT
cana-2550	78	29	as	as	SCONJ
cana-2550	78	30	opposed	oppose	VERB
cana-2550	78	31	to	to	ADP
cana-2550	78	32	the	the	DET
cana-2550	78	33	whole	whole	ADJ
cana-2550	78	34	dataset	dataset	NOUN
cana-2550	78	35	.	.	PUNCT
cana-2550	79	1	this	this	DET
cana-2550	79	2	method	method	NOUN
cana-2550	79	3	uses	use	VERB
cana-2550	79	4	randomness	randomness	NOUN
cana-2550	79	5	into	into	ADP
cana-2550	79	6	the	the	DET
cana-2550	79	7	optimization	optimization	NOUN
cana-2550	79	8	process	process	NOUN
cana-2550	79	9	,	,	PUNCT
cana-2550	79	10	which	which	PRON
cana-2550	79	11	helps	help	VERB
cana-2550	79	12	to	to	PART
cana-2550	79	13	improve	improve	VERB
cana-2550	79	14	convergence	convergence	NOUN
cana-2550	79	15	speed	speed	NOUN
cana-2550	79	16	and	and	CCONJ
cana-2550	79	17	which	which	PRON
cana-2550	79	18	in	in	ADP
cana-2550	79	19	turn	turn	NOUN
cana-2550	79	20	reduces	reduce	VERB
cana-2550	79	21	computational	computational	ADJ
cana-2550	79	22	requirements	requirement	NOUN
cana-2550	79	23	.	.	PUNCT
cana-2550	80	1	this	this	PRON
cana-2550	80	2	is	be	AUX
cana-2550	80	3	used	use	VERB
cana-2550	80	4	in	in	ADP
cana-2550	80	5	areas	area	NOUN
cana-2550	80	6	which	which	PRON
cana-2550	80	7	have	have	VERB
cana-2550	80	8	large	large	ADJ
cana-2550	80	9	datasets	dataset	NOUN
cana-2550	80	10	and	and	CCONJ
cana-2550	80	11	results	result	NOUN
cana-2550	80	12	in	in	ADP
cana-2550	80	13	improved	improved	ADJ
cana-2550	80	14	efficiency	efficiency	NOUN
cana-2550	80	15	.	.	PUNCT
cana-2550	81	1	the	the	DET
cana-2550	81	2	limitation	limitation	NOUN
cana-2550	81	3	of	of	ADP
cana-2550	81	4	sgd	sgd	PROPN
cana-2550	81	5	's	's	PART
cana-2550	81	6	stochastic	stochastic	ADJ
cana-2550	81	7	updates	update	NOUN
cana-2550	81	8	is	be	AUX
cana-2550	81	9	that	that	SCONJ
cana-2550	81	10	it	it	PRON
cana-2550	81	11	can	can	AUX
cana-2550	81	12	lead	lead	VERB
cana-2550	81	13	to	to	ADP
cana-2550	81	14	oscillations	oscillation	NOUN
cana-2550	81	15	near	near	ADP
cana-2550	81	16	the	the	DET
cana-2550	81	17	optimal	optimal	ADJ
cana-2550	81	18	solution	solution	NOUN
cana-2550	81	19	.	.	PUNCT
cana-2550	82	1	one	one	PRON
cana-2550	82	2	can	can	AUX
cana-2550	82	3	use	use	VERB
cana-2550	82	4	learning	learn	VERB
cana-2550	82	5	rate	rate	NOUN
cana-2550	82	6	decay	decay	NOUN
cana-2550	82	7	to	to	PART
cana-2550	82	8	overcome	overcome	VERB
cana-2550	82	9	this	this	DET
cana-2550	82	10	challenge	challenge	NOUN
cana-2550	82	11	.	.	PUNCT
cana-2550	83	1	sgd	sgd	PROPN
cana-2550	83	2	serves	serve	VERB
cana-2550	83	3	as	as	ADP
cana-2550	83	4	a	a	DET
cana-2550	83	5	dominant	dominant	ADJ
cana-2550	83	6	choice	choice	NOUN
cana-2550	83	7	in	in	ADP
cana-2550	83	8	deep	deep	ADJ
cana-2550	83	9	learning	learning	NOUN
cana-2550	83	10	due	due	ADP
cana-2550	83	11	to	to	ADP
cana-2550	83	12	its	its	PRON
cana-2550	83	13	effectiveness	effectiveness	NOUN
cana-2550	83	14	and	and	CCONJ
cana-2550	83	15	scalability	scalability	NOUN
cana-2550	83	16	.	.	PUNCT
cana-2550	84	1	θt+1	θt+1	X
cana-2550	85	1	=	=	PUNCT
cana-2550	85	2	θt	θt	PROPN
cana-2550	85	3	−	−	PROPN
cana-2550	85	4	η∇θℒ(θt	η∇θℒ(θt	NOUN
cana-2550	85	5	;	;	PUNCT
cana-2550	85	6	xi	xi	NUM
cana-2550	85	7	,	,	PUNCT
cana-2550	85	8	yi	yi	PROPN
cana-2550	85	9	)	)	PUNCT
cana-2550	86	1	𝜃𝑡	𝜃𝑡	CCONJ
cana-2550	86	2	represents	represent	VERB
cana-2550	86	3	the	the	DET
cana-2550	86	4	parameters	parameter	NOUN
cana-2550	86	5	at	at	ADP
cana-2550	86	6	iteration	iteration	NOUN
cana-2550	86	7	t	t	PROPN
cana-2550	86	8	𝜂	𝜂	NOUN
cana-2550	86	9	is	be	AUX
cana-2550	86	10	the	the	DET
cana-2550	86	11	learning	learning	NOUN
cana-2550	86	12	rate	rate	NOUN
cana-2550	86	13	∇𝜃ℒ(𝜃𝑡	∇𝜃ℒ(𝜃𝑡	PROPN
cana-2550	86	14	;	;	PUNCT
cana-2550	86	15	𝑥𝑖	𝑥𝑖	NUM
cana-2550	86	16	,	,	PUNCT
cana-2550	86	17	𝑦𝑖	𝑦𝑖	PROPN
cana-2550	86	18	)	)	PUNCT
cana-2550	86	19	is	be	AUX
cana-2550	86	20	the	the	DET
cana-2550	86	21	gradient	gradient	NOUN
cana-2550	86	22	of	of	ADP
cana-2550	86	23	the	the	DET
cana-2550	86	24	loss	loss	NOUN
cana-2550	86	25	function	function	NOUN
cana-2550	86	26	with	with	ADP
cana-2550	86	27	respect	respect	NOUN
cana-2550	86	28	to	to	ADP
cana-2550	86	29	the	the	DET
cana-2550	86	30	parameters	parameter	NOUN
cana-2550	86	31	at	at	ADP
cana-2550	86	32	iteration	iteration	PROPN
cana-2550	86	33	t	t	PROPN
cana-2550	86	34	,	,	PUNCT
cana-2550	86	35	computed	compute	VERB
cana-2550	86	36	using	use	VERB
cana-2550	86	37	a	a	DET
cana-2550	86	38	single	single	ADJ
cana-2550	86	39	data	data	NOUN
cana-2550	86	40	point	point	NOUN
cana-2550	86	41	𝑥𝑖	𝑥𝑖	PRON
cana-2550	86	42	,	,	PUNCT
cana-2550	86	43	𝑦𝑖	𝑦𝑖	PROPN
cana-2550	86	44	table	table	NOUN
cana-2550	86	45	1	1	NUM
cana-2550	86	46	:	:	PUNCT
cana-2550	86	47	comparison	comparison	NOUN
cana-2550	86	48	of	of	ADP
cana-2550	86	49	optimization	optimization	NOUN
cana-2550	86	50	methods	method	NOUN
cana-2550	86	51	on	on	ADP
cana-2550	86	52	convergence	convergence	NOUN
cana-2550	86	53	speed	speed	NOUN
cana-2550	86	54	optimization	optimization	NOUN
cana-2550	86	55	method	method	NOUN
cana-2550	86	56	learning	learn	VERB
cana-2550	86	57	rate	rate	NOUN
cana-2550	86	58	epochs	epoch	NOUN
cana-2550	86	59	to	to	PART
cana-2550	86	60	converge	converge	VERB
cana-2550	86	61	final	final	ADJ
cana-2550	86	62	training	training	NOUN
cana-2550	86	63	loss	loss	NOUN
cana-2550	86	64	final	final	ADJ
cana-2550	86	65	validation	validation	NOUN
cana-2550	86	66	loss	loss	NOUN
cana-2550	86	67	momentum	momentum	NOUN
cana-2550	86	68	based	base	VERB
cana-2550	86	69	sgd	sgd	PROPN
cana-2550	86	70	0.02	0.02	NUM
cana-2550	86	71	430	430	NUM
cana-2550	86	72	0.194	0.194	NUM
cana-2550	86	73	0.203	0.203	NUM
cana-2550	86	74	adagard	adagard	NOUN
cana-2550	86	75	0.01	0.01	NUM
cana-2550	86	76	550	550	NUM
cana-2550	86	77	0.204	0.204	NUM
cana-2550	86	78	0.213	0.213	NUM
cana-2550	86	79	gradient	gradient	ADJ
cana-2550	86	80	descent	descent	NOUN
cana-2550	86	81	0.01	0.01	NUM
cana-2550	86	82	495	495	NUM
cana-2550	86	83	0.218	0.218	NUM
cana-2550	86	84	0.219	0.219	NUM
cana-2550	86	85	adam	adam	PROPN
cana-2550	86	86	optimizer	optimizer	NOUN
cana-2550	86	87	0.001	0.001	NUM
cana-2550	86	88	410	410	NUM
cana-2550	86	89	0.180	0.180	NUM
cana-2550	86	90	0.190	0.190	NUM
cana-2550	86	91	sgd	sgd	X
cana-2550	86	92	0.01	0.01	NUM
cana-2550	86	93	560	560	NUM
cana-2550	86	94	0.210	0.210	NUM
cana-2550	86	95	0.225	0.225	NUM
cana-2550	86	96	description	description	NOUN
cana-2550	86	97	:	:	PUNCT
cana-2550	86	98	this	this	DET
cana-2550	86	99	table	table	NOUN
cana-2550	86	100	compares	compare	VERB
cana-2550	86	101	different	different	ADJ
cana-2550	86	102	optimization	optimization	NOUN
cana-2550	86	103	methods	method	NOUN
cana-2550	86	104	used	use	VERB
cana-2550	86	105	in	in	ADP
cana-2550	86	106	training	train	VERB
cana-2550	86	107	a	a	DET
cana-2550	86	108	neural	neural	ADJ
cana-2550	86	109	network	network	NOUN
cana-2550	86	110	.	.	PUNCT
cana-2550	87	1	the	the	DET
cana-2550	87	2	comparison	comparison	NOUN
cana-2550	87	3	focuses	focus	VERB
cana-2550	87	4	on	on	ADP
cana-2550	87	5	how	how	SCONJ
cana-2550	87	6	quickly	quickly	ADV
cana-2550	87	7	the	the	DET
cana-2550	87	8	methods	method	NOUN
cana-2550	87	9	converge	converge	VERB
cana-2550	87	10	(	(	PUNCT
cana-2550	87	11	measured	measure	VERB
cana-2550	87	12	in	in	ADP
cana-2550	87	13	epochs	epoch	NOUN
cana-2550	87	14	)	)	PUNCT
cana-2550	87	15	,	,	PUNCT
cana-2550	87	16	along	along	ADP
cana-2550	87	17	with	with	ADP
cana-2550	87	18	the	the	DET
cana-2550	87	19	final	final	ADJ
cana-2550	87	20	training	training	NOUN
cana-2550	87	21	and	and	CCONJ
cana-2550	87	22	validation	validation	NOUN
cana-2550	87	23	loss	loss	NOUN
cana-2550	87	24	values	value	NOUN
cana-2550	87	25	.	.	PUNCT
cana-2550	88	1	3.1.b	3.1.b	NUM
cana-2550	88	2	mini	mini	NOUN
cana-2550	88	3	-	-	NOUN
cana-2550	88	4	batch	batch	ADJ
cana-2550	88	5	gradient	gradient	ADJ
cana-2550	88	6	descent	descent	NOUN
cana-2550	88	7	:	:	PUNCT
cana-2550	88	8	it	it	PRON
cana-2550	88	9	is	be	AUX
cana-2550	88	10	also	also	ADV
cana-2550	88	11	one	one	NUM
cana-2550	88	12	of	of	ADP
cana-2550	88	13	the	the	DET
cana-2550	88	14	variations	variation	NOUN
cana-2550	88	15	of	of	ADP
cana-2550	88	16	gradient	gradient	ADJ
cana-2550	88	17	descent	descent	NOUN
cana-2550	88	18	known	know	VERB
cana-2550	88	19	as	as	ADP
cana-2550	88	20	mini	mini	NOUN
cana-2550	88	21	-	-	NOUN
cana-2550	88	22	batch	batch	ADJ
cana-2550	88	23	gradient	gradient	ADJ
cana-2550	88	24	descent	descent	NOUN
cana-2550	88	25	.	.	PUNCT
cana-2550	89	1	it	it	PRON
cana-2550	89	2	utilises	utilise	VERB
cana-2550	89	3	a	a	DET
cana-2550	89	4	tiny	tiny	ADJ
cana-2550	89	5	,	,	PUNCT
cana-2550	89	6	random	random	ADJ
cana-2550	89	7	subset	subset	NOUN
cana-2550	89	8	of	of	ADP
cana-2550	89	9	the	the	DET
cana-2550	89	10	training	training	NOUN
cana-2550	89	11	data	datum	NOUN
cana-2550	89	12	—	—	PUNCT
cana-2550	89	13	called	call	VERB
cana-2550	89	14	as	as	ADP
cana-2550	89	15	a	a	DET
cana-2550	89	16	minibatch	minibatch	NOUN
cana-2550	89	17	—	—	PUNCT
cana-2550	89	18	instead	instead	ADV
cana-2550	89	19	of	of	ADP
cana-2550	89	20	the	the	DET
cana-2550	89	21	full	full	ADJ
cana-2550	89	22	dataset	dataset	NOUN
cana-2550	89	23	or	or	CCONJ
cana-2550	89	24	a	a	DET
cana-2550	89	25	single	single	ADJ
cana-2550	89	26	data	data	NOUN
cana-2550	89	27	point	point	NOUN
cana-2550	89	28	to	to	PART
cana-2550	89	29	calculate	calculate	VERB
cana-2550	89	30	the	the	DET
cana-2550	89	31	gradient	gradient	NOUN
cana-2550	89	32	.	.	PUNCT
cana-2550	90	1	this	this	DET
cana-2550	90	2	method	method	NOUN
cana-2550	90	3	balances	balance	VERB
cana-2550	90	4	the	the	DET
cana-2550	90	5	faster	fast	ADJ
cana-2550	90	6	convergence	convergence	NOUN
cana-2550	90	7	of	of	ADP
cana-2550	90	8	stochastic	stochastic	ADJ
cana-2550	90	9	gradient	gradient	ADJ
cana-2550	90	10	descent	descent	NOUN
cana-2550	90	11	(	(	PUNCT
cana-2550	90	12	sgd	sgd	PROPN
cana-2550	90	13	)	)	PUNCT
cana-2550	90	14	with	with	ADP
cana-2550	90	15	the	the	DET
cana-2550	90	16	computational	computational	ADJ
cana-2550	90	17	economy	economy	NOUN
cana-2550	90	18	of	of	ADP
cana-2550	90	19	fullbatch	fullbatch	NOUN
cana-2550	90	20	gradient	gradient	ADJ
cana-2550	90	21	descent	descent	NOUN
cana-2550	90	22	.	.	PUNCT
cana-2550	91	1	the	the	DET
cana-2550	91	2	samples	sample	NOUN
cana-2550	91	3	used	use	VERB
cana-2550	91	4	in	in	ADP
cana-2550	91	5	this	this	DET
cana-2550	91	6	method	method	NOUN
cana-2550	91	7	depends	depend	VERB
cana-2550	91	8	on	on	ADP
cana-2550	91	9	the	the	DET
cana-2550	91	10	size	size	NOUN
cana-2550	91	11	of	of	ADP
cana-2550	91	12	the	the	DET
cana-2550	91	13	model	model	NOUN
cana-2550	91	14	and	and	CCONJ
cana-2550	91	15	the	the	DET
cana-2550	91	16	communications	communication	NOUN
cana-2550	91	17	on	on	ADP
cana-2550	91	18	applied	apply	VERB
cana-2550	91	19	nonlinear	nonlinear	ADJ
cana-2550	91	20	analysis	analysis	NOUN
cana-2550	91	21	issn	issn	NOUN
cana-2550	91	22	:	:	PUNCT
cana-2550	91	23	1074	1074	NUM
cana-2550	91	24	-	-	PUNCT
cana-2550	91	25	133x	133x	NUM
cana-2550	91	26	vol	vol	NOUN
cana-2550	91	27	32	32	NUM
cana-2550	91	28	no	no	NOUN
cana-2550	91	29	.	.	PUNCT
cana-2550	92	1	3s	3s	NUM
cana-2550	92	2	(	(	PUNCT
cana-2550	92	3	2025	2025	NUM
cana-2550	92	4	)	)	PUNCT
cana-2550	92	5	72	72	NUM
cana-2550	93	1	https://internationalpubls.com	https://internationalpubls.com	X
cana-2550	93	2	dataset	dataset	NOUN
cana-2550	93	3	.	.	PUNCT
cana-2550	94	1	mini	mini	ADJ
cana-2550	94	2	batches	batch	NOUN
cana-2550	94	3	reduce	reduce	VERB
cana-2550	94	4	noise	noise	NOUN
cana-2550	94	5	in	in	ADP
cana-2550	94	6	the	the	DET
cana-2550	94	7	optimization	optimization	NOUN
cana-2550	94	8	process	process	NOUN
cana-2550	94	9	compared	compare	VERB
cana-2550	94	10	to	to	ADP
cana-2550	94	11	sgd	sgd	PROPN
cana-2550	94	12	,	,	PUNCT
cana-2550	94	13	enabling	enable	VERB
cana-2550	94	14	more	more	ADV
cana-2550	94	15	consistent	consistent	ADJ
cana-2550	94	16	and	and	CCONJ
cana-2550	94	17	seamless	seamless	ADJ
cana-2550	94	18	updates	update	NOUN
cana-2550	94	19	.	.	PUNCT
cana-2550	95	1	the	the	DET
cana-2550	95	2	calculation	calculation	NOUN
cana-2550	95	3	efficiency	efficiency	NOUN
cana-2550	95	4	of	of	ADP
cana-2550	95	5	this	this	DET
cana-2550	95	6	method	method	NOUN
cana-2550	95	7	is	be	AUX
cana-2550	95	8	greater	great	ADJ
cana-2550	95	9	than	than	ADP
cana-2550	95	10	the	the	DET
cana-2550	95	11	previous	previous	ADJ
cana-2550	95	12	methods	method	NOUN
cana-2550	95	13	.	.	PUNCT
cana-2550	96	1	in	in	ADP
cana-2550	96	2	deep	deep	ADJ
cana-2550	96	3	learning	learning	NOUN
cana-2550	96	4	models	model	NOUN
cana-2550	96	5	,	,	PUNCT
cana-2550	96	6	mini	mini	NOUN
cana-2550	96	7	-	-	NOUN
cana-2550	96	8	batch	batch	ADJ
cana-2550	96	9	method	method	NOUN
cana-2550	96	10	proves	prove	VERB
cana-2550	96	11	to	to	PART
cana-2550	96	12	be	be	AUX
cana-2550	96	13	more	more	ADV
cana-2550	96	14	efficient	efficient	ADJ
cana-2550	96	15	in	in	ADP
cana-2550	96	16	terms	term	NOUN
cana-2550	96	17	of	of	ADP
cana-2550	96	18	computational	computational	ADJ
cana-2550	96	19	time	time	NOUN
cana-2550	96	20	and	and	CCONJ
cana-2550	96	21	speed	speed	NOUN
cana-2550	96	22	for	for	ADP
cana-2550	96	23	large	large	ADJ
cana-2550	96	24	datasets	dataset	NOUN
cana-2550	96	25	.	.	PUNCT
cana-2550	97	1	additionally	additionally	ADV
cana-2550	97	2	,	,	PUNCT
cana-2550	97	3	it	it	PRON
cana-2550	97	4	can	can	AUX
cana-2550	97	5	speed	speed	VERB
cana-2550	97	6	up	up	ADP
cana-2550	97	7	training	training	NOUN
cana-2550	97	8	by	by	ADP
cana-2550	97	9	utilizing	utilize	VERB
cana-2550	97	10	gpu	gpu	NOUN
cana-2550	97	11	acceleration	acceleration	NOUN
cana-2550	97	12	and	and	CCONJ
cana-2550	97	13	parallelization	parallelization	NOUN
cana-2550	97	14	.	.	PUNCT
cana-2550	98	1	the	the	DET
cana-2550	98	2	size	size	NOUN
cana-2550	98	3	of	of	ADP
cana-2550	98	4	the	the	DET
cana-2550	98	5	mini	mini	ADJ
cana-2550	98	6	batch	batch	NOUN
cana-2550	98	7	is	be	AUX
cana-2550	98	8	crucial	crucial	ADJ
cana-2550	98	9	.	.	PUNCT
cana-2550	99	1	if	if	SCONJ
cana-2550	99	2	the	the	DET
cana-2550	99	3	mini	mini	ADJ
cana-2550	99	4	batch	batch	NOUN
cana-2550	99	5	used	use	VERB
cana-2550	99	6	is	be	AUX
cana-2550	99	7	too	too	ADV
cana-2550	99	8	tiny	tiny	ADJ
cana-2550	99	9	then	then	ADV
cana-2550	99	10	it	it	PRON
cana-2550	99	11	will	will	AUX
cana-2550	99	12	produce	produce	VERB
cana-2550	99	13	noisy	noisy	ADJ
cana-2550	99	14	gradients	gradient	NOUN
cana-2550	99	15	,	,	PUNCT
cana-2550	99	16	and	and	CCONJ
cana-2550	99	17	if	if	SCONJ
cana-2550	99	18	it	it	PRON
cana-2550	99	19	is	be	AUX
cana-2550	99	20	too	too	ADV
cana-2550	99	21	large	large	ADJ
cana-2550	99	22	then	then	ADV
cana-2550	99	23	it	it	PRON
cana-2550	99	24	will	will	AUX
cana-2550	99	25	reduce	reduce	VERB
cana-2550	99	26	the	the	DET
cana-2550	99	27	efficiency	efficiency	NOUN
cana-2550	99	28	improvements	improvement	NOUN
cana-2550	99	29	.	.	PUNCT
cana-2550	100	1	figure	figure	VERB
cana-2550	100	2	1	1	NUM
cana-2550	100	3	comparison	comparison	NOUN
cana-2550	100	4	of	of	ADP
cana-2550	100	5	optimization	optimization	NOUN
cana-2550	100	6	methods	method	NOUN
cana-2550	100	7	for	for	ADP
cana-2550	100	8	convergence	convergence	NOUN
cana-2550	100	9	speed	speed	NOUN
cana-2550	100	10	3.2	3.2	NUM
cana-2550	100	11	momentum	momentum	NOUN
cana-2550	100	12	-	-	PUNCT
cana-2550	100	13	based	base	VERB
cana-2550	100	14	methods	method	NOUN
cana-2550	100	15	:	:	PUNCT
cana-2550	100	16	momentum	momentum	NOUN
cana-2550	100	17	-	-	PUNCT
cana-2550	100	18	based	base	VERB
cana-2550	100	19	optimization	optimization	NOUN
cana-2550	100	20	methods	method	NOUN
cana-2550	100	21	are	be	AUX
cana-2550	100	22	strategies	strategy	NOUN
cana-2550	100	23	used	use	VERB
cana-2550	100	24	to	to	PART
cana-2550	100	25	enhance	enhance	VERB
cana-2550	100	26	gradient	gradient	ADJ
cana-2550	100	27	descent	descent	NOUN
cana-2550	100	28	by	by	ADP
cana-2550	100	29	integrating	integrate	VERB
cana-2550	100	30	past	past	ADJ
cana-2550	100	31	gradients	gradient	NOUN
cana-2550	100	32	into	into	ADP
cana-2550	100	33	the	the	DET
cana-2550	100	34	current	current	ADJ
cana-2550	100	35	update	update	NOUN
cana-2550	100	36	,	,	PUNCT
cana-2550	100	37	aiming	aim	VERB
cana-2550	100	38	to	to	PART
cana-2550	100	39	emphasize	emphasize	VERB
cana-2550	100	40	previous	previous	ADJ
cana-2550	100	41	gradients	gradient	NOUN
cana-2550	100	42	for	for	ADP
cana-2550	100	43	smoother	smooth	ADJ
cana-2550	100	44	adjustments	adjustment	NOUN
cana-2550	100	45	and	and	CCONJ
cana-2550	100	46	faster	fast	ADJ
cana-2550	100	47	convergence	convergence	NOUN
cana-2550	100	48	.	.	PUNCT
cana-2550	101	1	this	this	PRON
cana-2550	101	2	involves	involve	VERB
cana-2550	101	3	the	the	DET
cana-2550	101	4	concept	concept	NOUN
cana-2550	101	5	of	of	ADP
cana-2550	101	6	adjusting	adjust	VERB
cana-2550	101	7	the	the	DET
cana-2550	101	8	update	update	NOUN
cana-2550	101	9	rule	rule	NOUN
cana-2550	101	10	by	by	ADP
cana-2550	101	11	introducing	introduce	VERB
cana-2550	101	12	a	a	DET
cana-2550	101	13	\velocity\	\velocity\	NOUN
cana-2550	101	14	component	component	NOUN
cana-2550	101	15	,	,	PUNCT
cana-2550	101	16	which	which	PRON
cana-2550	101	17	represents	represent	VERB
cana-2550	101	18	a	a	DET
cana-2550	101	19	weighted	weighted	ADJ
cana-2550	101	20	average	average	NOUN
cana-2550	101	21	of	of	ADP
cana-2550	101	22	prior	prior	ADJ
cana-2550	101	23	gradients	gradient	NOUN
cana-2550	101	24	.	.	PUNCT
cana-2550	102	1	momentum	momentum	NOUN
cana-2550	102	2	plays	play	VERB
cana-2550	102	3	a	a	DET
cana-2550	102	4	crucial	crucial	ADJ
cana-2550	102	5	role	role	NOUN
cana-2550	102	6	in	in	ADP
cana-2550	102	7	advancing	advance	VERB
cana-2550	102	8	convergence	convergence	NOUN
cana-2550	102	9	toward	toward	ADP
cana-2550	102	10	the	the	DET
cana-2550	102	11	relevant	relevant	ADJ
cana-2550	102	12	direction	direction	NOUN
cana-2550	102	13	and	and	CCONJ
cana-2550	102	14	mitigating	mitigate	VERB
cana-2550	102	15	oscillations	oscillation	NOUN
cana-2550	102	16	,	,	PUNCT
cana-2550	102	17	particularly	particularly	ADV
cana-2550	102	18	in	in	ADP
cana-2550	102	19	areas	area	NOUN
cana-2550	102	20	with	with	ADP
cana-2550	102	21	varying	vary	VERB
cana-2550	102	22	gradients	gradient	NOUN
cana-2550	102	23	.	.	PUNCT
cana-2550	103	1	in	in	ADP
cana-2550	103	2	deep	deep	ADJ
cana-2550	103	3	learning	learning	NOUN
cana-2550	103	4	,	,	PUNCT
cana-2550	103	5	this	this	DET
cana-2550	103	6	strategy	strategy	NOUN
cana-2550	103	7	proves	prove	VERB
cana-2550	103	8	to	to	PART
cana-2550	103	9	be	be	AUX
cana-2550	103	10	beneficial	beneficial	ADJ
cana-2550	103	11	as	as	SCONJ
cana-2550	103	12	it	it	PRON
cana-2550	103	13	addresses	address	VERB
cana-2550	103	14	the	the	DET
cana-2550	103	15	issue	issue	NOUN
cana-2550	103	16	of	of	ADP
cana-2550	103	17	slow	slow	ADJ
cana-2550	103	18	convergence	convergence	NOUN
cana-2550	103	19	caused	cause	VERB
cana-2550	103	20	by	by	ADP
cana-2550	103	21	flat	flat	ADJ
cana-2550	103	22	areas	area	NOUN
cana-2550	103	23	or	or	CCONJ
cana-2550	103	24	steep	steep	ADJ
cana-2550	103	25	gradients	gradient	NOUN
cana-2550	103	26	within	within	ADP
cana-2550	103	27	the	the	DET
cana-2550	103	28	loss	loss	NOUN
cana-2550	103	29	landscape	landscape	NOUN
cana-2550	103	30	directions	direction	NOUN
cana-2550	103	31	.	.	PUNCT
cana-2550	104	1	𝑣𝑡	𝑣𝑡	ADP
cana-2550	104	2	=	=	PUNCT
cana-2550	105	1	β𝑣𝑡−1	β𝑣𝑡−1	PUNCT
cana-2550	105	2	+	+	PUNCT
cana-2550	105	3	(	(	PUNCT
cana-2550	105	4	1	1	NUM
cana-2550	105	5	−	−	NOUN
cana-2550	105	6	β)∇θℒ(θ𝑡	β)∇θℒ(θ𝑡	NOUN
cana-2550	105	7	)	)	PUNCT
cana-2550	105	8	]	]	PUNCT
cana-2550	105	9	θ𝑡+1	θ𝑡+1	NOUN
cana-2550	105	10	=	=	SYM
cana-2550	105	11	θ𝑡	θ𝑡	PROPN
cana-2550	105	12	−	−	PROPN
cana-2550	105	13	η𝑣𝑡	η𝑣𝑡	VERB
cana-2550	105	14	𝑣𝑡	𝑣𝑡	NOUN
cana-2550	105	15	is	be	AUX
cana-2550	105	16	the	the	DET
cana-2550	105	17	velocity	velocity	NOUN
cana-2550	105	18	at	at	ADP
cana-2550	105	19	time	time	NOUN
cana-2550	105	20	t.	t.	PROPN
cana-2550	105	21	𝛽	𝛽	PROPN
cana-2550	105	22	is	be	AUX
cana-2550	105	23	the	the	DET
cana-2550	105	24	momentum	momentum	NOUN
cana-2550	105	25	term	term	NOUN
cana-2550	105	26	.	.	PUNCT
cana-2550	106	1	∇𝜃ℒ(𝜃𝑡	∇𝜃ℒ(𝜃𝑡	NOUN
cana-2550	106	2	)	)	PUNCT
cana-2550	106	3	]	]	PUNCT
cana-2550	106	4	is	be	AUX
cana-2550	106	5	the	the	DET
cana-2550	106	6	gradient	gradient	NOUN
cana-2550	106	7	of	of	ADP
cana-2550	106	8	the	the	DET
cana-2550	106	9	loss	loss	NOUN
cana-2550	106	10	function	function	NOUN
cana-2550	106	11	with	with	ADP
cana-2550	106	12	respect	respect	NOUN
cana-2550	106	13	to	to	ADP
cana-2550	106	14	the	the	DET
cana-2550	106	15	parameters	parameter	NOUN
cana-2550	106	16	at	at	ADP
cana-2550	106	17	time	time	NOUN
cana-2550	106	18	t.	t.	PROPN
cana-2550	106	19	𝜃𝑡+1	𝜃𝑡+1	NUM
cana-2550	106	20	is	be	AUX
cana-2550	106	21	the	the	DET
cana-2550	106	22	updated	update	VERB
cana-2550	106	23	parameter	parameter	NOUN
cana-2550	106	24	at	at	ADP
cana-2550	106	25	time	time	NOUN
cana-2550	106	26	t+1	t+1	PROPN
cana-2550	106	27	𝜂	𝜂	PROPN
cana-2550	106	28	is	be	AUX
cana-2550	106	29	the	the	DET
cana-2550	106	30	learning	learning	NOUN
cana-2550	106	31	rate	rate	NOUN
cana-2550	106	32	4	4	NUM
cana-2550	106	33	.	.	PUNCT
cana-2550	106	34	advanced	advanced	ADJ
cana-2550	106	35	optimization	optimization	NOUN
cana-2550	106	36	technique	technique	NOUN
cana-2550	106	37	:	:	PUNCT
cana-2550	106	38	advanced	advanced	ADJ
cana-2550	106	39	optimization	optimization	NOUN
cana-2550	106	40	techniques	technique	NOUN
cana-2550	106	41	are	be	AUX
cana-2550	106	42	introduced	introduce	VERB
cana-2550	106	43	in	in	ADP
cana-2550	106	44	place	place	NOUN
cana-2550	106	45	of	of	ADP
cana-2550	106	46	traditional	traditional	ADJ
cana-2550	106	47	gradient	gradient	ADJ
cana-2550	106	48	descent	descent	NOUN
cana-2550	106	49	methods	method	NOUN
cana-2550	106	50	to	to	PART
cana-2550	106	51	improve	improve	VERB
cana-2550	106	52	the	the	DET
cana-2550	106	53	speed	speed	NOUN
cana-2550	106	54	and	and	CCONJ
cana-2550	106	55	quality	quality	NOUN
cana-2550	106	56	of	of	ADP
cana-2550	106	57	convergence	convergence	NOUN
cana-2550	106	58	in	in	ADP
cana-2550	106	59	complex	complex	ADJ
cana-2550	106	60	machine	machine	NOUN
cana-2550	106	61	learning	learning	NOUN
cana-2550	106	62	tasks	task	NOUN
cana-2550	106	63	.	.	PUNCT
cana-2550	107	1	4.1	4.1	NUM
cana-2550	107	2	adagrad	adagrad	ADJ
cana-2550	107	3	:	:	PUNCT
cana-2550	107	4	adagrad	adagrad	PROPN
cana-2550	107	5	stands	stand	VERB
cana-2550	107	6	for	for	ADP
cana-2550	107	7	adaptive	adaptive	ADJ
cana-2550	107	8	gradient	gradient	ADJ
cana-2550	107	9	algorithm	algorithm	NOUN
cana-2550	107	10	.	.	PUNCT
cana-2550	108	1	it	it	PRON
cana-2550	108	2	is	be	AUX
cana-2550	108	3	an	an	DET
cana-2550	108	4	optimization	optimization	NOUN
cana-2550	108	5	technique	technique	NOUN
cana-2550	108	6	that	that	PRON
cana-2550	108	7	converts	convert	VERB
cana-2550	108	8	the	the	DET
cana-2550	108	9	learning	learning	NOUN
cana-2550	108	10	rate	rate	NOUN
cana-2550	108	11	for	for	ADP
cana-2550	108	12	individual	individual	ADJ
cana-2550	108	13	parameters	parameter	NOUN
cana-2550	108	14	by	by	ADP
cana-2550	108	15	considering	consider	VERB
cana-2550	108	16	their	their	PRON
cana-2550	108	17	previous	previous	ADJ
cana-2550	108	18	gradients	gradient	NOUN
cana-2550	108	19	.	.	PUNCT
cana-2550	109	1	this	this	DET
cana-2550	109	2	method	method	NOUN
cana-2550	109	3	is	be	AUX
cana-2550	109	4	used	use	VERB
cana-2550	109	5	to	to	PART
cana-2550	109	6	handle	handle	VERB
cana-2550	109	7	thin	thin	ADJ
cana-2550	109	8	data	datum	NOUN
cana-2550	109	9	by	by	ADP
cana-2550	109	10	making	make	VERB
cana-2550	109	11	adjustments	adjustment	NOUN
cana-2550	109	12	in	in	ADP
cana-2550	109	13	the	the	DET
cana-2550	109	14	learning	learning	NOUN
cana-2550	109	15	rate	rate	NOUN
cana-2550	109	16	.	.	PUNCT
cana-2550	110	1	the	the	DET
cana-2550	110	2	adjustment	adjustment	NOUN
cana-2550	110	3	is	be	AUX
cana-2550	110	4	made	make	VERB
cana-2550	110	5	by	by	ADP
cana-2550	110	6	scaling	scale	VERB
cana-2550	110	7	the	the	DET
cana-2550	110	8	learning	learning	NOUN
cana-2550	110	9	rate	rate	NOUN
cana-2550	110	10	inversely	inversely	ADV
cana-2550	110	11	proportional	proportional	ADJ
cana-2550	110	12	to	to	ADP
cana-2550	110	13	the	the	DET
cana-2550	110	14	square	square	ADJ
cana-2550	110	15	root	root	NOUN
cana-2550	110	16	of	of	ADP
cana-2550	110	17	the	the	DET
cana-2550	110	18	cumulative	cumulative	ADJ
cana-2550	110	19	sum	sum	NOUN
cana-2550	110	20	of	of	ADP
cana-2550	110	21	squared	square	VERB
cana-2550	110	22	past	past	ADJ
cana-2550	110	23	gradients	gradient	NOUN
cana-2550	110	24	.	.	PUNCT
cana-2550	111	1	subsequently	subsequently	ADV
cana-2550	111	2	,	,	PUNCT
cana-2550	111	3	parameters	parameter	NOUN
cana-2550	111	4	updated	update	VERB
cana-2550	111	5	frequently	frequently	ADV
cana-2550	111	6	receive	receive	VERB
cana-2550	111	7	smaller	small	ADJ
cana-2550	111	8	updates	update	NOUN
cana-2550	111	9	,	,	PUNCT
cana-2550	111	10	while	while	SCONJ
cana-2550	111	11	those	those	DET
cana-2550	111	12	communications	communication	NOUN
cana-2550	111	13	on	on	ADP
cana-2550	111	14	applied	apply	VERB
cana-2550	111	15	nonlinear	nonlinear	ADJ
cana-2550	111	16	analysis	analysis	NOUN
cana-2550	111	17	issn	issn	NOUN
cana-2550	111	18	:	:	PUNCT
cana-2550	111	19	1074	1074	NUM
cana-2550	111	20	-	-	PUNCT
cana-2550	111	21	133x	133x	NUM
cana-2550	111	22	vol	vol	NOUN
cana-2550	111	23	32	32	NUM
cana-2550	111	24	no	no	NOUN
cana-2550	111	25	.	.	PUNCT
cana-2550	112	1	3s	3s	NUM
cana-2550	112	2	(	(	PUNCT
cana-2550	112	3	2025	2025	NUM
cana-2550	112	4	)	)	PUNCT
cana-2550	112	5	73	73	NUM
cana-2550	112	6	https://internationalpubls.com	https://internationalpubls.com	NUM
cana-2550	112	7	updated	update	VERB
cana-2550	112	8	less	less	ADV
cana-2550	112	9	frequently	frequently	ADV
cana-2550	112	10	receive	receive	VERB
cana-2550	112	11	more	more	ADV
cana-2550	112	12	substantial	substantial	ADJ
cana-2550	112	13	updates	update	NOUN
cana-2550	112	14	.	.	PUNCT
cana-2550	113	1	this	this	DET
cana-2550	113	2	feature	feature	NOUN
cana-2550	113	3	of	of	ADP
cana-2550	113	4	the	the	DET
cana-2550	113	5	algorithm	algorithm	NOUN
cana-2550	113	6	helps	help	VERB
cana-2550	113	7	to	to	PART
cana-2550	113	8	accelerate	accelerate	VERB
cana-2550	113	9	the	the	DET
cana-2550	113	10	convergence	convergence	NOUN
cana-2550	113	11	of	of	ADP
cana-2550	113	12	the	the	DET
cana-2550	113	13	algorithm	algorithm	NOUN
cana-2550	113	14	,	,	PUNCT
cana-2550	113	15	particularly	particularly	ADV
cana-2550	113	16	in	in	ADP
cana-2550	113	17	situations	situation	NOUN
cana-2550	113	18	involving	involve	VERB
cana-2550	113	19	sparse	sparse	ADJ
cana-2550	113	20	data	datum	NOUN
cana-2550	113	21	or	or	CCONJ
cana-2550	113	22	features	feature	NOUN
cana-2550	113	23	.	.	PUNCT
cana-2550	114	1	the	the	DET
cana-2550	114	2	update	update	NOUN
cana-2550	114	3	formula	formula	NOUN
cana-2550	114	4	for	for	ADP
cana-2550	114	5	adagrad	adagrad	NOUN
cana-2550	114	6	is	be	AUX
cana-2550	114	7	as	as	SCONJ
cana-2550	114	8	follows	follow	VERB
cana-2550	114	9	:	:	PUNCT
cana-2550	114	10	θ𝑡+1	θ𝑡+1	NOUN
cana-2550	114	11	=	=	SYM
cana-2550	114	12	θ𝑡	θ𝑡	PROPN
cana-2550	114	13	−	−	PROPN
cana-2550	114	14	η	η	X
cana-2550	114	15	√𝐺𝑡	√𝐺𝑡	PROPN
cana-2550	114	16	+	+	PROPN
cana-2550	114	17	ϵ	ϵ	X
cana-2550	114	18	∇θℒ(θ𝑡	∇θℒ(θ𝑡	PROPN
cana-2550	114	19	)	)	PUNCT
cana-2550	114	20	𝜃𝑡	𝜃𝑡	ADV
cana-2550	114	21	is	be	AUX
cana-2550	114	22	the	the	DET
cana-2550	114	23	parameter	parameter	NOUN
cana-2550	114	24	at	at	ADP
cana-2550	114	25	time	time	NOUN
cana-2550	114	26	t	t	PROPN
cana-2550	114	27	,	,	PUNCT
cana-2550	114	28	𝜂	𝜂	PROPN
cana-2550	114	29	is	be	AUX
cana-2550	114	30	the	the	DET
cana-2550	114	31	learning	learning	NOUN
cana-2550	114	32	rate	rate	NOUN
cana-2550	114	33	,	,	PUNCT
cana-2550	114	34	𝐺𝑡	𝐺𝑡	PROPN
cana-2550	114	35	is	be	AUX
cana-2550	114	36	the	the	DET
cana-2550	114	37	sum	sum	NOUN
cana-2550	114	38	of	of	ADP
cana-2550	114	39	the	the	DET
cana-2550	114	40	squared	square	VERB
cana-2550	114	41	gradients	gradient	NOUN
cana-2550	114	42	up	up	ADP
cana-2550	114	43	to	to	ADP
cana-2550	114	44	time	time	NOUN
cana-2550	114	45	ttt	ttt	PROPN
cana-2550	114	46	,	,	PUNCT
cana-2550	114	47	𝜖	𝜖	PROPN
cana-2550	114	48	is	be	AUX
cana-2550	114	49	a	a	DET
cana-2550	114	50	small	small	ADJ
cana-2550	114	51	constant	constant	NOUN
cana-2550	114	52	to	to	PART
cana-2550	114	53	prevent	prevent	VERB
cana-2550	114	54	division	division	NOUN
cana-2550	114	55	by	by	ADP
cana-2550	114	56	zero	zero	NUM
cana-2550	114	57	,	,	PUNCT
cana-2550	114	58	∇θl(θt	∇θl(θt	NOUN
cana-2550	114	59	)	)	PUNCT
cana-2550	114	60	is	be	AUX
cana-2550	114	61	the	the	DET
cana-2550	114	62	gradient	gradient	NOUN
cana-2550	114	63	of	of	ADP
cana-2550	114	64	the	the	DET
cana-2550	114	65	loss	loss	NOUN
cana-2550	114	66	function	function	NOUN
cana-2550	114	67	with	with	ADP
cana-2550	114	68	respect	respect	NOUN
cana-2550	114	69	to	to	ADP
cana-2550	114	70	the	the	DET
cana-2550	114	71	parameters	parameter	NOUN
cana-2550	114	72	at	at	ADP
cana-2550	114	73	time	time	NOUN
cana-2550	114	74	ttt	ttt	PROPN
cana-2550	114	75	.	.	PUNCT
cana-2550	114	76	figure	figure	NOUN
cana-2550	114	77	2	2	NUM
cana-2550	114	78	performance	performance	NOUN
cana-2550	114	79	with	with	ADP
cana-2550	114	80	different	different	ADJ
cana-2550	114	81	regularization	regularization	NOUN
cana-2550	114	82	techniques	technique	VERB
cana-2550	114	83	4.2	4.2	NUM
cana-2550	114	84	newtons	newton	NOUN
cana-2550	114	85	method	method	NOUN
cana-2550	114	86	:	:	PUNCT
cana-2550	114	87	newton	newton	PROPN
cana-2550	114	88	’s	’s	PART
cana-2550	114	89	method	method	NOUN
cana-2550	114	90	is	be	AUX
cana-2550	114	91	an	an	DET
cana-2550	114	92	optimization	optimization	NOUN
cana-2550	114	93	technique	technique	NOUN
cana-2550	114	94	that	that	PRON
cana-2550	114	95	uses	use	VERB
cana-2550	114	96	second	second	ADJ
cana-2550	114	97	-	-	PUNCT
cana-2550	114	98	order	order	NOUN
cana-2550	114	99	information	information	NOUN
cana-2550	114	100	to	to	PART
cana-2550	114	101	find	find	VERB
cana-2550	114	102	the	the	DET
cana-2550	114	103	roots	root	NOUN
cana-2550	114	104	of	of	ADP
cana-2550	114	105	a	a	DET
cana-2550	114	106	function	function	NOUN
cana-2550	114	107	or	or	CCONJ
cana-2550	114	108	minimize	minimize	VERB
cana-2550	114	109	an	an	DET
cana-2550	114	110	objective	objective	ADJ
cana-2550	114	111	function	function	NOUN
cana-2550	114	112	.	.	PUNCT
cana-2550	115	1	newton	newton	PROPN
cana-2550	115	2	’s	’s	PART
cana-2550	115	3	method	method	PROPN
cana-2550	115	4	implements	implement	VERB
cana-2550	115	5	the	the	DET
cana-2550	115	6	second	second	ADJ
cana-2550	115	7	derivative	derivative	NOUN
cana-2550	115	8	(	(	PUNCT
cana-2550	115	9	hessian	hessian	ADJ
cana-2550	115	10	matrix	matrix	NOUN
cana-2550	115	11	)	)	PUNCT
cana-2550	115	12	to	to	PART
cana-2550	115	13	adjust	adjust	VERB
cana-2550	115	14	the	the	DET
cana-2550	115	15	update	update	NOUN
cana-2550	115	16	whereas	whereas	SCONJ
cana-2550	115	17	gradient	gradient	ADJ
cana-2550	115	18	based	base	VERB
cana-2550	115	19	methods	method	NOUN
cana-2550	115	20	depends	depend	VERB
cana-2550	115	21	on	on	ADP
cana-2550	115	22	first	first	ADJ
cana-2550	115	23	order	order	NOUN
cana-2550	115	24	deriatives	deriative	NOUN
cana-2550	115	25	.	.	PUNCT
cana-2550	116	1	the	the	DET
cana-2550	116	2	update	update	NOUN
cana-2550	116	3	rule	rule	NOUN
cana-2550	116	4	for	for	ADP
cana-2550	116	5	newton	newton	PROPN
cana-2550	116	6	’s	’s	PART
cana-2550	116	7	method	method	NOUN
cana-2550	116	8	is	be	AUX
cana-2550	116	9	:	:	PUNCT
cana-2550	116	10	θ𝑡+1	θ𝑡+1	NOUN
cana-2550	116	11	=	=	SYM
cana-2550	116	12	θ𝑡	θ𝑡	PROPN
cana-2550	116	13	−	−	PROPN
cana-2550	116	14	η𝐻−1∇θℒ(θ𝑡	η𝐻−1∇θℒ(θ𝑡	NOUN
cana-2550	116	15	)	)	PUNCT
cana-2550	116	16	𝜂	𝜂	NOUN
cana-2550	116	17	is	be	AUX
cana-2550	116	18	the	the	DET
cana-2550	116	19	learning	learning	NOUN
cana-2550	116	20	rate	rate	NOUN
cana-2550	116	21	𝐻	𝐻	PROPN
cana-2550	116	22	is	be	AUX
cana-2550	116	23	the	the	DET
cana-2550	116	24	hessian	hessian	ADJ
cana-2550	116	25	matrix	matrix	NOUN
cana-2550	116	26	,	,	PUNCT
cana-2550	116	27	∇𝜃ℒ(𝜃𝑡	∇𝜃ℒ(𝜃𝑡	PROPN
cana-2550	116	28	)	)	PUNCT
cana-2550	116	29	is	be	AUX
cana-2550	116	30	the	the	DET
cana-2550	116	31	gradient	gradient	NOUN
cana-2550	116	32	of	of	ADP
cana-2550	116	33	loss	loss	NOUN
cana-2550	116	34	function	function	NOUN
cana-2550	116	35	when	when	SCONJ
cana-2550	116	36	employing	employ	VERB
cana-2550	116	37	curve	curve	NOUN
cana-2550	116	38	,	,	PUNCT
cana-2550	116	39	newton	newton	PROPN
cana-2550	116	40	’s	’s	PART
cana-2550	116	41	method	method	NOUN
cana-2550	116	42	have	have	VERB
cana-2550	116	43	other	other	ADJ
cana-2550	116	44	benefits	benefit	NOUN
cana-2550	116	45	like	like	SCONJ
cana-2550	116	46	they	they	PRON
cana-2550	116	47	have	have	VERB
cana-2550	116	48	the	the	DET
cana-2550	116	49	potential	potential	NOUN
cana-2550	116	50	to	to	PART
cana-2550	116	51	achieve	achieve	VERB
cana-2550	116	52	quicker	quick	ADJ
cana-2550	116	53	convergence	convergence	NOUN
cana-2550	116	54	compared	compare	VERB
cana-2550	116	55	to	to	ADP
cana-2550	116	56	first	first	ADJ
cana-2550	116	57	-	-	PUNCT
cana-2550	116	58	order	order	NOUN
cana-2550	116	59	methods	method	NOUN
cana-2550	116	60	,	,	PUNCT
cana-2550	116	61	particularly	particularly	ADV
cana-2550	116	62	evident	evident	ADJ
cana-2550	116	63	in	in	ADP
cana-2550	116	64	convex	convex	ADJ
cana-2550	116	65	issues	issue	NOUN
cana-2550	116	66	.	.	PUNCT
cana-2550	117	1	the	the	DET
cana-2550	117	2	computation	computation	NOUN
cana-2550	117	3	of	of	ADP
cana-2550	117	4	the	the	DET
cana-2550	117	5	hessian	hessian	ADJ
cana-2550	117	6	presents	present	VERB
cana-2550	117	7	challenges	challenge	NOUN
cana-2550	117	8	in	in	ADP
cana-2550	117	9	high	high	ADJ
cana-2550	117	10	-	-	PUNCT
cana-2550	117	11	dimensional	dimensional	ADJ
cana-2550	117	12	scenarios	scenario	NOUN
cana-2550	117	13	due	due	ADP
cana-2550	117	14	to	to	ADP
cana-2550	117	15	its	its	PRON
cana-2550	117	16	computational	computational	ADJ
cana-2550	117	17	intensity	intensity	NOUN
cana-2550	117	18	,	,	PUNCT
cana-2550	117	19	rendering	render	VERB
cana-2550	117	20	it	it	PRON
cana-2550	117	21	impractical	impractical	ADJ
cana-2550	117	22	for	for	ADP
cana-2550	117	23	extensive	extensive	ADJ
cana-2550	117	24	datasets	dataset	NOUN
cana-2550	117	25	.	.	PUNCT
cana-2550	118	1	newton	newton	PROPN
cana-2550	118	2	’s	’s	PART
cana-2550	118	3	method	method	NOUN
cana-2550	118	4	proves	prove	VERB
cana-2550	118	5	especially	especially	ADV
cana-2550	118	6	beneficial	beneficial	ADJ
cana-2550	118	7	when	when	SCONJ
cana-2550	118	8	dealing	deal	VERB
cana-2550	118	9	with	with	ADP
cana-2550	118	10	smooth	smooth	ADJ
cana-2550	118	11	and	and	CCONJ
cana-2550	118	12	well	well	ADV
cana-2550	118	13	-	-	PUNCT
cana-2550	118	14	behaved	behave	VERB
cana-2550	118	15	objective	objective	ADJ
cana-2550	118	16	functions	function	NOUN
cana-2550	118	17	,	,	PUNCT
cana-2550	118	18	facilitating	facilitate	VERB
cana-2550	118	19	more	more	ADV
cana-2550	118	20	precise	precise	ADJ
cana-2550	118	21	strides	stride	NOUN
cana-2550	118	22	towards	towards	ADP
cana-2550	118	23	the	the	DET
cana-2550	118	24	optimal	optimal	ADJ
cana-2550	118	25	solution	solution	NOUN
cana-2550	118	26	.	.	PUNCT
cana-2550	119	1	however	however	ADV
cana-2550	119	2	,	,	PUNCT
cana-2550	119	3	when	when	SCONJ
cana-2550	119	4	challenged	challenge	VERB
cana-2550	119	5	with	with	ADP
cana-2550	119	6	non	non	ADJ
cana-2550	119	7	-	-	ADJ
cana-2550	119	8	convex	convex	ADJ
cana-2550	119	9	problems	problem	NOUN
cana-2550	119	10	,	,	PUNCT
cana-2550	119	11	it	it	PRON
cana-2550	119	12	may	may	AUX
cana-2550	119	13	encounter	encounter	VERB
cana-2550	119	14	obstacles	obstacle	NOUN
cana-2550	119	15	such	such	ADJ
cana-2550	119	16	as	as	ADP
cana-2550	119	17	frame	frame	NOUN
cana-2550	119	18	in	in	ADP
cana-2550	119	19	load	load	NOUN
cana-2550	119	20	sites	site	NOUN
cana-2550	119	21	.	.	PUNCT
cana-2550	120	1	table	table	NOUN
cana-2550	120	2	2	2	NUM
cana-2550	120	3	:	:	PUNCT
cana-2550	120	4	performance	performance	NOUN
cana-2550	120	5	with	with	ADP
cana-2550	120	6	different	different	ADJ
cana-2550	120	7	regularization	regularization	NOUN
cana-2550	120	8	techniques	technique	NOUN
cana-2550	120	9	regularization	regularization	NOUN
cana-2550	120	10	type	type	NOUN
cana-2550	120	11	regularization	regularization	NOUN
cana-2550	120	12	strength	strength	NOUN
cana-2550	120	13	training	training	NOUN
cana-2550	120	14	loss	loss	NOUN
cana-2550	120	15	validation	validation	NOUN
cana-2550	120	16	loss	loss	NOUN
cana-2550	120	17	final	final	ADJ
cana-2550	120	18	accuracy	accuracy	NOUN
cana-2550	120	19	(	(	PUNCT
cana-2550	120	20	%	%	INTJ
cana-2550	120	21	)	)	PUNCT
cana-2550	120	22	l2	l2	NOUN
cana-2550	120	23	regularization	regularization	NOUN
cana-2550	120	24	0.001	0.001	NUM
cana-2550	120	25	0.219	0.219	NUM
cana-2550	120	26	0.223	0.223	NUM
cana-2550	120	27	90.2	90.2	NUM
cana-2550	120	28	l1	l1	PROPN
cana-2550	120	29	regularization	regularization	NOUN
cana-2550	120	30	0.001	0.001	NUM
cana-2550	120	31	0.225	0.225	NUM
cana-2550	120	32	0.225	0.225	NUM
cana-2550	120	33	89.0	89.0	NUM
cana-2550	120	34	no	no	DET
cana-2550	120	35	regularization	regularization	NOUN
cana-2550	120	36	0.0	0.0	NUM
cana-2550	120	37	0.250	0.250	NUM
cana-2550	120	38	0.262	0.262	NUM
cana-2550	120	39	86.0	86.0	NUM
cana-2550	120	40	dropout	dropout	NOUN
cana-2550	120	41	(	(	PUNCT
cana-2550	120	42	0.5	0.5	NUM
cana-2550	120	43	)	)	PUNCT
cana-2550	120	44	n	n	CCONJ
cana-2550	120	45	/	/	SYM
cana-2550	120	46	a	a	DET
cana-2550	120	47	0.235	0.235	NUM
cana-2550	120	48	0.245	0.245	NUM
cana-2550	120	49	88.9	88.9	NUM
cana-2550	120	50	communications	communication	NOUN
cana-2550	120	51	on	on	ADP
cana-2550	120	52	applied	apply	VERB
cana-2550	120	53	nonlinear	nonlinear	ADJ
cana-2550	120	54	analysis	analysis	NOUN
cana-2550	120	55	issn	issn	NOUN
cana-2550	120	56	:	:	PUNCT
cana-2550	120	57	1074	1074	NUM
cana-2550	120	58	-	-	PUNCT
cana-2550	120	59	133x	133x	NUM
cana-2550	120	60	vol	vol	NOUN
cana-2550	120	61	32	32	NUM
cana-2550	121	1	no	no	NOUN
cana-2550	121	2	.	.	PUNCT
cana-2550	122	1	3s	3s	NUM
cana-2550	122	2	(	(	PUNCT
cana-2550	122	3	2025	2025	NUM
cana-2550	122	4	)	)	PUNCT
cana-2550	122	5	74	74	NUM
cana-2550	122	6	https://internationalpubls.com	https://internationalpubls.com	X
cana-2550	122	7	description	description	NOUN
cana-2550	122	8	:	:	PUNCT
cana-2550	122	9	this	this	DET
cana-2550	122	10	table	table	NOUN
cana-2550	122	11	compares	compare	VERB
cana-2550	122	12	the	the	DET
cana-2550	122	13	impact	impact	NOUN
cana-2550	122	14	of	of	ADP
cana-2550	122	15	different	different	ADJ
cana-2550	122	16	regularization	regularization	NOUN
cana-2550	122	17	techniques	technique	NOUN
cana-2550	122	18	(	(	PUNCT
cana-2550	122	19	l2	l2	NOUN
cana-2550	122	20	,	,	PUNCT
cana-2550	122	21	l1	l1	PROPN
cana-2550	122	22	,	,	PUNCT
cana-2550	122	23	and	and	CCONJ
cana-2550	122	24	dropout	dropout	NOUN
cana-2550	122	25	)	)	PUNCT
cana-2550	122	26	on	on	ADP
cana-2550	122	27	model	model	NOUN
cana-2550	122	28	performance	performance	NOUN
cana-2550	122	29	.	.	PUNCT
cana-2550	123	1	the	the	DET
cana-2550	123	2	results	result	NOUN
cana-2550	123	3	demonstrate	demonstrate	VERB
cana-2550	123	4	the	the	DET
cana-2550	123	5	effectiveness	effectiveness	NOUN
cana-2550	123	6	of	of	ADP
cana-2550	123	7	l2	l2	NOUN
cana-2550	123	8	regularization	regularization	NOUN
cana-2550	123	9	in	in	ADP
cana-2550	123	10	improving	improve	VERB
cana-2550	123	11	both	both	PRON
cana-2550	123	12	training	training	NOUN
cana-2550	123	13	and	and	CCONJ
cana-2550	123	14	validation	validation	NOUN
cana-2550	123	15	loss	loss	NOUN
cana-2550	123	16	while	while	SCONJ
cana-2550	123	17	maintaining	maintain	VERB
cana-2550	123	18	high	high	ADJ
cana-2550	123	19	accuracy	accuracy	NOUN
cana-2550	123	20	.	.	PUNCT
cana-2550	124	1	5	5	X
cana-2550	124	2	.	.	NUM
cana-2550	124	3	proposed	propose	VERB
cana-2550	124	4	optimization	optimization	NOUN
cana-2550	124	5	method	method	NOUN
cana-2550	124	6	for	for	ADP
cana-2550	124	7	neural	neural	ADJ
cana-2550	124	8	networks	network	NOUN
cana-2550	124	9	:	:	PUNCT
cana-2550	124	10	in	in	ADP
cana-2550	124	11	this	this	DET
cana-2550	124	12	segment	segment	NOUN
cana-2550	124	13	,	,	PUNCT
cana-2550	124	14	we	we	PRON
cana-2550	124	15	introduce	introduce	VERB
cana-2550	124	16	a	a	DET
cana-2550	124	17	fresh	fresh	ADJ
cana-2550	124	18	optimization	optimization	NOUN
cana-2550	124	19	technique	technique	NOUN
cana-2550	124	20	specifically	specifically	ADV
cana-2550	124	21	shaped	shape	VERB
cana-2550	124	22	to	to	PART
cana-2550	124	23	boost	boost	VERB
cana-2550	124	24	the	the	DET
cana-2550	124	25	efficiency	efficiency	NOUN
cana-2550	124	26	and	and	CCONJ
cana-2550	124	27	speed	speed	NOUN
cana-2550	124	28	of	of	ADP
cana-2550	124	29	convergence	convergence	NOUN
cana-2550	124	30	when	when	SCONJ
cana-2550	124	31	training	train	VERB
cana-2550	124	32	deep	deep	ADJ
cana-2550	124	33	neural	neural	ADJ
cana-2550	124	34	networks	network	NOUN
cana-2550	124	35	.	.	PUNCT
cana-2550	125	1	in	in	ADP
cana-2550	125	2	this	this	DET
cana-2550	125	3	method	method	NOUN
cana-2550	125	4	,	,	PUNCT
cana-2550	125	5	elements	element	NOUN
cana-2550	125	6	of	of	ADP
cana-2550	125	7	momentum	momentum	NOUN
cana-2550	125	8	-	-	PUNCT
cana-2550	125	9	based	base	VERB
cana-2550	125	10	strategies	strategy	NOUN
cana-2550	125	11	,	,	PUNCT
cana-2550	125	12	adaptive	adaptive	ADJ
cana-2550	125	13	learning	learning	NOUN
cana-2550	125	14	rates	rate	NOUN
cana-2550	125	15	,	,	PUNCT
cana-2550	125	16	and	and	CCONJ
cana-2550	125	17	second	second	ADJ
cana-2550	125	18	-	-	PUNCT
cana-2550	125	19	order	order	NOUN
cana-2550	125	20	optimization	optimization	NOUN
cana-2550	125	21	methods	method	NOUN
cana-2550	125	22	are	be	AUX
cana-2550	125	23	merged	merge	VERB
cana-2550	125	24	to	to	PART
cana-2550	125	25	offer	offer	VERB
cana-2550	125	26	a	a	DET
cana-2550	125	27	corresponding	corresponding	ADJ
cana-2550	125	28	and	and	CCONJ
cana-2550	125	29	computationally	computationally	ADV
cana-2550	125	30	efficient	efficient	ADJ
cana-2550	125	31	optimization	optimization	NOUN
cana-2550	125	32	strategy	strategy	NOUN
cana-2550	125	33	.	.	PUNCT
cana-2550	126	1	the	the	DET
cana-2550	126	2	essential	essential	ADJ
cana-2550	126	3	concept	concept	NOUN
cana-2550	126	4	involves	involve	VERB
cana-2550	126	5	connecting	connect	VERB
cana-2550	126	6	the	the	DET
cana-2550	126	7	advantages	advantage	NOUN
cana-2550	126	8	of	of	ADP
cana-2550	126	9	both	both	DET
cana-2550	126	10	firstand	firstand	NOUN
cana-2550	126	11	second	second	ADJ
cana-2550	126	12	-	-	PUNCT
cana-2550	126	13	order	order	NOUN
cana-2550	126	14	methodologies	methodology	NOUN
cana-2550	126	15	while	while	SCONJ
cana-2550	126	16	addressing	address	VERB
cana-2550	126	17	their	their	PRON
cana-2550	126	18	respective	respective	ADJ
cana-2550	126	19	drawbacks	drawback	NOUN
cana-2550	126	20	,	,	PUNCT
cana-2550	126	21	such	such	ADJ
cana-2550	126	22	as	as	ADP
cana-2550	126	23	the	the	DET
cana-2550	126	24	computational	computational	ADJ
cana-2550	126	25	intensity	intensity	NOUN
cana-2550	126	26	associated	associate	VERB
cana-2550	126	27	with	with	ADP
cana-2550	126	28	second	second	ADJ
cana-2550	126	29	-	-	PUNCT
cana-2550	126	30	order	order	NOUN
cana-2550	126	31	techniques	technique	NOUN
cana-2550	126	32	and	and	CCONJ
cana-2550	126	33	the	the	DET
cana-2550	126	34	inherent	inherent	ADJ
cana-2550	126	35	oscillations	oscillation	NOUN
cana-2550	126	36	in	in	ADP
cana-2550	126	37	first	first	ADJ
cana-2550	126	38	-	-	PUNCT
cana-2550	126	39	order	order	NOUN
cana-2550	126	40	methods	method	NOUN
cana-2550	126	41	.	.	PUNCT
cana-2550	127	1	5.1	5.1	NUM
cana-2550	127	2	hybrid	hybrid	NOUN
cana-2550	127	3	gradient	gradient	NOUN
cana-2550	127	4	and	and	CCONJ
cana-2550	127	5	momentumbased	momentumbased	ADJ
cana-2550	127	6	updates	update	NOUN
cana-2550	127	7	:	:	PUNCT
cana-2550	127	8	-this	-this	ADJ
cana-2550	127	9	method	method	NOUN
cana-2550	127	10	initiates	initiate	VERB
cana-2550	127	11	with	with	ADP
cana-2550	127	12	a	a	DET
cana-2550	127	13	standard	standard	ADJ
cana-2550	127	14	gradient	gradient	ADJ
cana-2550	127	15	descent	descent	NOUN
cana-2550	127	16	update	update	NOUN
cana-2550	127	17	,	,	PUNCT
cana-2550	127	18	and	and	CCONJ
cana-2550	127	19	later	later	ADV
cana-2550	127	20	introduces	introduce	VERB
cana-2550	127	21	a	a	DET
cana-2550	127	22	momentum	momentum	NOUN
cana-2550	127	23	term	term	NOUN
cana-2550	127	24	to	to	PART
cana-2550	127	25	help	help	VERB
cana-2550	127	26	accelerate	accelerate	VERB
cana-2550	127	27	convergence	convergence	NOUN
cana-2550	127	28	.	.	PUNCT
cana-2550	128	1	this	this	DET
cana-2550	128	2	term	term	NOUN
cana-2550	128	3	is	be	AUX
cana-2550	128	4	weighted	weight	VERB
cana-2550	128	5	by	by	ADP
cana-2550	128	6	a	a	DET
cana-2550	128	7	factor	factor	NOUN
cana-2550	128	8	β	β	NOUN
cana-2550	128	9	,	,	PUNCT
cana-2550	128	10	which	which	PRON
cana-2550	128	11	allows	allow	VERB
cana-2550	128	12	previous	previous	ADJ
cana-2550	128	13	gradients	gradient	NOUN
cana-2550	128	14	to	to	PART
cana-2550	128	15	have	have	VERB
cana-2550	128	16	a	a	DET
cana-2550	128	17	lasting	last	VERB
cana-2550	128	18	influence	influence	NOUN
cana-2550	128	19	on	on	ADP
cana-2550	128	20	the	the	DET
cana-2550	128	21	current	current	ADJ
cana-2550	128	22	update	update	NOUN
cana-2550	128	23	.	.	PUNCT
cana-2550	129	1	the	the	DET
cana-2550	129	2	update	update	NOUN
cana-2550	129	3	rule	rule	NOUN
cana-2550	129	4	is	be	AUX
cana-2550	129	5	given	give	VERB
cana-2550	129	6	by	by	ADP
cana-2550	129	7	:	:	PUNCT
cana-2550	129	8	𝑣𝑡+1	𝑣𝑡+1	PRON
cana-2550	129	9	=	=	SYM
cana-2550	129	10	β𝑣𝑡	β𝑣𝑡	NOUN
cana-2550	129	11	+	+	CCONJ
cana-2550	129	12	(	(	PUNCT
cana-2550	129	13	1	1	NUM
cana-2550	129	14	−	−	NOUN
cana-2550	129	15	β)∇θℒ(θ𝑡	β)∇θℒ(θ𝑡	NOUN
cana-2550	129	16	)	)	PUNCT
cana-2550	129	17	θ𝑡+1	θ𝑡+1	NOUN
cana-2550	129	18	=	=	SYM
cana-2550	129	19	θ𝑡	θ𝑡	PROPN
cana-2550	129	20	−	−	PROPN
cana-2550	129	21	η𝑣𝑡+1	η𝑣𝑡+1	NOUN
cana-2550	129	22	𝜃𝑡	𝜃𝑡	ADV
cana-2550	129	23	is	be	AUX
cana-2550	129	24	the	the	DET
cana-2550	129	25	parameter	parameter	NOUN
cana-2550	129	26	at	at	ADP
cana-2550	129	27	iteration	iteration	PROPN
cana-2550	129	28	t	t	PROPN
cana-2550	129	29	,	,	PUNCT
cana-2550	129	30	𝑣𝑡+1	𝑣𝑡+1	X
cana-2550	129	31	is	be	AUX
cana-2550	129	32	the	the	DET
cana-2550	129	33	momentum	momentum	NOUN
cana-2550	129	34	term	term	NOUN
cana-2550	129	35	at	at	ADP
cana-2550	129	36	iteration	iteration	NOUN
cana-2550	129	37	t+1	t+1	PROPN
cana-2550	129	38	𝜂	𝜂	PROPN
cana-2550	129	39	is	be	AUX
cana-2550	129	40	the	the	DET
cana-2550	129	41	learning	learning	NOUN
cana-2550	129	42	rate	rate	NOUN
cana-2550	129	43	,	,	PUNCT
cana-2550	129	44	𝜃𝑡+1	𝜃𝑡+1	PRON
cana-2550	129	45	is	be	AUX
cana-2550	129	46	the	the	DET
cana-2550	129	47	updated	update	VERB
cana-2550	129	48	parameter	parameter	NOUN
cana-2550	129	49	at	at	ADP
cana-2550	129	50	iteration	iteration	NOUN
cana-2550	129	51	t+1	t+1	PRON
cana-2550	129	52	figure	figure	VERB
cana-2550	129	53	3	3	NUM
cana-2550	129	54	final	final	ADJ
cana-2550	129	55	accuracy	accuracy	NOUN
cana-2550	129	56	5.2	5.2	NUM
cana-2550	129	57	adaptive	adaptive	ADJ
cana-2550	129	58	learning	learning	NOUN
cana-2550	129	59	rates	rate	NOUN
cana-2550	129	60	with	with	ADP
cana-2550	129	61	second	second	ADJ
cana-2550	129	62	-	-	PUNCT
cana-2550	129	63	order	order	NOUN
cana-2550	129	64	information	information	NOUN
cana-2550	129	65	:	:	PUNCT
cana-2550	129	66	to	to	PART
cana-2550	129	67	further	far	ADV
cana-2550	129	68	improve	improve	VERB
cana-2550	129	69	the	the	DET
cana-2550	129	70	optimization	optimization	NOUN
cana-2550	129	71	process	process	NOUN
cana-2550	129	72	,	,	PUNCT
cana-2550	129	73	we	we	PRON
cana-2550	129	74	merge	merge	VERB
cana-2550	129	75	an	an	DET
cana-2550	129	76	adaptive	adaptive	ADJ
cana-2550	129	77	learning	learning	NOUN
cana-2550	129	78	rate	rate	NOUN
cana-2550	129	79	mechanism	mechanism	NOUN
cana-2550	129	80	which	which	PRON
cana-2550	129	81	will	will	AUX
cana-2550	129	82	adjust	adjust	VERB
cana-2550	129	83	the	the	DET
cana-2550	129	84	learning	learning	NOUN
cana-2550	129	85	rate	rate	NOUN
cana-2550	129	86	for	for	ADP
cana-2550	129	87	each	each	DET
cana-2550	129	88	parameter	parameter	NOUN
cana-2550	129	89	which	which	PRON
cana-2550	129	90	depends	depend	VERB
cana-2550	129	91	on	on	ADP
cana-2550	129	92	the	the	DET
cana-2550	129	93	degree	degree	NOUN
cana-2550	129	94	of	of	ADP
cana-2550	129	95	its	its	PRON
cana-2550	129	96	gradient	gradient	NOUN
cana-2550	129	97	.	.	PUNCT
cana-2550	130	1	we	we	PRON
cana-2550	130	2	will	will	AUX
cana-2550	130	3	also	also	ADV
cana-2550	130	4	introduce	introduce	VERB
cana-2550	130	5	secondorder	secondorder	ADJ
cana-2550	130	6	information	information	NOUN
cana-2550	130	7	by	by	ADP
cana-2550	130	8	utilizing	utilize	VERB
cana-2550	130	9	the	the	DET
cana-2550	130	10	hessian	hessian	ADJ
cana-2550	130	11	matrix	matrix	NOUN
cana-2550	130	12	to	to	PART
cana-2550	130	13	guide	guide	VERB
cana-2550	130	14	the	the	DET
cana-2550	130	15	parameter	parameter	NOUN
cana-2550	130	16	updates	update	VERB
cana-2550	130	17	.	.	PUNCT
cana-2550	131	1	this	this	DET
cana-2550	131	2	approach	approach	NOUN
cana-2550	131	3	combines	combine	VERB
cana-2550	131	4	the	the	DET
cana-2550	131	5	efficiency	efficiency	NOUN
cana-2550	131	6	of	of	ADP
cana-2550	131	7	first	first	ADJ
cana-2550	131	8	-	-	PUNCT
cana-2550	131	9	order	order	NOUN
cana-2550	131	10	methods	method	NOUN
cana-2550	131	11	with	with	ADP
cana-2550	131	12	the	the	DET
cana-2550	131	13	more	more	ADV
cana-2550	131	14	accurate	accurate	ADJ
cana-2550	131	15	and	and	CCONJ
cana-2550	131	16	rapid	rapid	ADJ
cana-2550	131	17	convergence	convergence	NOUN
cana-2550	131	18	typically	typically	ADV
cana-2550	131	19	associated	associate	VERB
cana-2550	131	20	with	with	ADP
cana-2550	131	21	second	second	ADJ
cana-2550	131	22	-	-	PUNCT
cana-2550	131	23	order	order	NOUN
cana-2550	131	24	methods	method	NOUN
cana-2550	131	25	like	like	ADP
cana-2550	131	26	newton	newton	PROPN
cana-2550	131	27	’s	’s	PART
cana-2550	131	28	method	method	NOUN
cana-2550	131	29	.	.	PUNCT
cana-2550	132	1	the	the	DET
cana-2550	132	2	update	update	NOUN
cana-2550	132	3	rule	rule	NOUN
cana-2550	132	4	incorporating	incorporate	VERB
cana-2550	132	5	both	both	CCONJ
cana-2550	132	6	the	the	DET
cana-2550	132	7	adaptive	adaptive	ADJ
cana-2550	132	8	learning	learning	NOUN
cana-2550	132	9	rate	rate	NOUN
cana-2550	132	10	and	and	CCONJ
cana-2550	132	11	second	second	ADJ
cana-2550	132	12	-	-	PUNCT
cana-2550	132	13	order	order	NOUN
cana-2550	132	14	information	information	NOUN
cana-2550	132	15	is	be	AUX
cana-2550	132	16	:	:	PUNCT
cana-2550	132	17	θt+1	θt+1	NUM
cana-2550	132	18	=	=	PUNCT
cana-2550	132	19	θt	θt	PROPN
cana-2550	132	20	−	−	PROPN
cana-2550	132	21	ηh−1∇θℒ(θt	ηh−1∇θℒ(θt	PROPN
cana-2550	132	22	)	)	PUNCT
cana-2550	132	23	𝜃𝑡	𝜃𝑡	ADV
cana-2550	132	24	is	be	AUX
cana-2550	132	25	the	the	DET
cana-2550	132	26	parameter	parameter	NOUN
cana-2550	132	27	at	at	ADP
cana-2550	132	28	iteration	iteration	PROPN
cana-2550	132	29	t	t	PROPN
cana-2550	132	30	,	,	PUNCT
cana-2550	132	31	𝜂	𝜂	PROPN
cana-2550	132	32	is	be	AUX
cana-2550	132	33	the	the	DET
cana-2550	132	34	learning	learning	NOUN
cana-2550	132	35	rate	rate	NOUN
cana-2550	132	36	,	,	PUNCT
cana-2550	132	37	𝐻−1	𝐻−1	PROPN
cana-2550	132	38	is	be	AUX
cana-2550	132	39	the	the	DET
cana-2550	132	40	inverse	inverse	NOUN
cana-2550	132	41	of	of	ADP
cana-2550	132	42	the	the	DET
cana-2550	132	43	hessian	hessian	ADJ
cana-2550	132	44	matrix	matrix	NOUN
cana-2550	132	45	(	(	PUNCT
cana-2550	132	46	second	second	ADJ
cana-2550	132	47	derivative	derivative	NOUN
cana-2550	132	48	of	of	ADP
cana-2550	132	49	the	the	DET
cana-2550	132	50	loss	loss	NOUN
cana-2550	132	51	function	function	NOUN
cana-2550	132	52	)	)	PUNCT
cana-2550	132	53	,	,	PUNCT
cana-2550	132	54	∇𝜃ℒ(𝜃𝑡	∇𝜃ℒ(𝜃𝑡	NOUN
cana-2550	132	55	)	)	PUNCT
cana-2550	132	56	is	be	AUX
cana-2550	132	57	the	the	DET
cana-2550	132	58	gradient	gradient	NOUN
cana-2550	132	59	of	of	ADP
cana-2550	132	60	the	the	DET
cana-2550	132	61	loss	loss	NOUN
cana-2550	132	62	function	function	NOUN
cana-2550	132	63	at	at	ADP
cana-2550	132	64	iteration	iteration	NOUN
cana-2550	132	65	t.	t.	NOUN
cana-2550	132	66	communications	communication	NOUN
cana-2550	132	67	on	on	ADP
cana-2550	132	68	applied	apply	VERB
cana-2550	132	69	nonlinear	nonlinear	ADJ
cana-2550	132	70	analysis	analysis	NOUN
cana-2550	132	71	issn	issn	NOUN
cana-2550	132	72	:	:	PUNCT
cana-2550	132	73	1074	1074	NUM
cana-2550	132	74	-	-	PUNCT
cana-2550	132	75	133x	133x	NUM
cana-2550	132	76	vol	vol	NOUN
cana-2550	132	77	32	32	NUM
cana-2550	132	78	no	no	NOUN
cana-2550	132	79	.	.	PUNCT
cana-2550	133	1	3s	3s	NUM
cana-2550	133	2	(	(	PUNCT
cana-2550	133	3	2025	2025	NUM
cana-2550	133	4	)	)	PUNCT
cana-2550	133	5	75	75	NUM
cana-2550	133	6	https://internationalpubls.com	https://internationalpubls.com	X
cana-2550	133	7	5.3	5.3	NUM
cana-2550	133	8	handling	handle	VERB
cana-2550	133	9	large	large	ADJ
cana-2550	133	10	-	-	PUNCT
cana-2550	133	11	scale	scale	NOUN
cana-2550	133	12	datasets	dataset	NOUN
cana-2550	133	13	with	with	ADP
cana-2550	133	14	stochastic	stochastic	ADJ
cana-2550	133	15	gradient	gradient	ADJ
cana-2550	133	16	descent	descent	NOUN
cana-2550	133	17	(	(	PUNCT
cana-2550	133	18	sgd	sgd	PROPN
cana-2550	133	19	):	):	PUNCT
cana-2550	133	20	since	since	SCONJ
cana-2550	133	21	training	train	VERB
cana-2550	133	22	deep	deep	ADJ
cana-2550	133	23	neural	neural	ADJ
cana-2550	133	24	networks	network	NOUN
cana-2550	133	25	often	often	ADV
cana-2550	133	26	involves	involve	VERB
cana-2550	133	27	large	large	ADJ
cana-2550	133	28	datasets	dataset	NOUN
cana-2550	133	29	,	,	PUNCT
cana-2550	133	30	the	the	DET
cana-2550	133	31	method	method	NOUN
cana-2550	133	32	integrates	integrate	VERB
cana-2550	133	33	a	a	DET
cana-2550	133	34	mini	mini	NOUN
cana-2550	133	35	-	-	NOUN
cana-2550	133	36	batch	batch	ADJ
cana-2550	133	37	stochastic	stochastic	ADJ
cana-2550	133	38	gradient	gradient	ADJ
cana-2550	133	39	descent	descent	NOUN
cana-2550	133	40	(	(	PUNCT
cana-2550	133	41	sgd	sgd	NOUN
cana-2550	133	42	)	)	PUNCT
cana-2550	133	43	approach	approach	NOUN
cana-2550	133	44	.	.	PUNCT
cana-2550	134	1	mini	mini	ADJ
cana-2550	134	2	-	-	NOUN
cana-2550	134	3	batch	batch	NOUN
cana-2550	134	4	sgd	sgd	NOUN
cana-2550	134	5	updates	update	VERB
cana-2550	134	6	the	the	DET
cana-2550	134	7	model	model	NOUN
cana-2550	134	8	parameters	parameter	NOUN
cana-2550	134	9	using	use	VERB
cana-2550	134	10	a	a	DET
cana-2550	134	11	random	random	ADJ
cana-2550	134	12	subset	subset	NOUN
cana-2550	134	13	of	of	ADP
cana-2550	134	14	the	the	DET
cana-2550	134	15	data	datum	NOUN
cana-2550	134	16	,	,	PUNCT
cana-2550	134	17	making	make	VERB
cana-2550	134	18	the	the	DET
cana-2550	134	19	optimization	optimization	NOUN
cana-2550	134	20	process	process	NOUN
cana-2550	134	21	faster	fast	ADV
cana-2550	134	22	and	and	CCONJ
cana-2550	134	23	more	more	ADV
cana-2550	134	24	computationally	computationally	ADV
cana-2550	134	25	feasible	feasible	ADJ
cana-2550	134	26	.	.	PUNCT
cana-2550	135	1	this	this	PRON
cana-2550	135	2	allows	allow	VERB
cana-2550	135	3	the	the	DET
cana-2550	135	4	method	method	NOUN
cana-2550	135	5	to	to	PART
cana-2550	135	6	scale	scale	VERB
cana-2550	135	7	to	to	ADP
cana-2550	135	8	large	large	ADJ
cana-2550	135	9	datasets	dataset	NOUN
cana-2550	135	10	without	without	ADP
cana-2550	135	11	sacrificing	sacrifice	VERB
cana-2550	135	12	too	too	ADV
cana-2550	135	13	much	much	ADJ
cana-2550	135	14	accuracy	accuracy	NOUN
cana-2550	135	15	.	.	PUNCT
cana-2550	136	1	the	the	DET
cana-2550	136	2	mini	mini	ADJ
cana-2550	136	3	-	-	NOUN
cana-2550	136	4	batch	batch	ADJ
cana-2550	136	5	update	update	NOUN
cana-2550	136	6	rule	rule	NOUN
cana-2550	136	7	is	be	AUX
cana-2550	136	8	given	give	VERB
cana-2550	136	9	by	by	ADP
cana-2550	136	10	:	:	PUNCT
cana-2550	136	11	θ𝑡+1	θ𝑡+1	NOUN
cana-2550	136	12	=	=	SYM
cana-2550	136	13	θ𝑡	θ𝑡	PROPN
cana-2550	136	14	−	−	PROPN
cana-2550	136	15	η∇θℒ(θ𝑡	η∇θℒ(θ𝑡	PROPN
cana-2550	136	16	;	;	PUNCT
cana-2550	136	17	𝑋batch	𝑋batch	PROPN
cana-2550	136	18	,	,	PUNCT
cana-2550	136	19	𝑌batch	𝑌batch	PROPN
cana-2550	136	20	)	)	PUNCT
cana-2550	136	21	where	where	SCONJ
cana-2550	136	22	𝑋batch	𝑋batch	PROPN
cana-2550	136	23	,	,	PUNCT
cana-2550	136	24	𝑌batch	𝑌batch	PROPN
cana-2550	136	25	represents	represent	VERB
cana-2550	136	26	a	a	DET
cana-2550	136	27	mini	mini	NOUN
cana-2550	136	28	-	-	NOUN
cana-2550	136	29	batch	batch	NOUN
cana-2550	136	30	of	of	ADP
cana-2550	136	31	training	training	NOUN
cana-2550	136	32	data	datum	NOUN
cana-2550	136	33	,	,	PUNCT
cana-2550	136	34	and	and	CCONJ
cana-2550	136	35	the	the	DET
cana-2550	136	36	gradient	gradient	NOUN
cana-2550	136	37	is	be	AUX
cana-2550	136	38	computed	compute	VERB
cana-2550	136	39	using	use	VERB
cana-2550	136	40	this	this	DET
cana-2550	136	41	batch	batch	NOUN
cana-2550	136	42	.	.	PUNCT
cana-2550	137	1	5.4	5.4	NUM
cana-2550	137	2	regularization	regularization	NOUN
cana-2550	137	3	and	and	CCONJ
cana-2550	137	4	adaptive	adaptive	ADJ
cana-2550	137	5	stopping	stopping	NOUN
cana-2550	137	6	criterion	criterion	NOUN
cana-2550	137	7	:	:	PUNCT
cana-2550	137	8	we	we	PRON
cana-2550	137	9	integrate	integrate	VERB
cana-2550	137	10	regularization	regularization	NOUN
cana-2550	137	11	methods	method	NOUN
cana-2550	137	12	like	like	ADP
cana-2550	137	13	l2	l2	NOUN
cana-2550	137	14	regularization	regularization	NOUN
cana-2550	137	15	into	into	ADP
cana-2550	137	16	the	the	DET
cana-2550	137	17	optimization	optimization	NOUN
cana-2550	137	18	process	process	NOUN
cana-2550	137	19	.	.	PUNCT
cana-2550	138	1	this	this	PRON
cana-2550	138	2	will	will	AUX
cana-2550	138	3	help	help	VERB
cana-2550	138	4	to	to	PART
cana-2550	138	5	prevent	prevent	VERB
cana-2550	138	6	overfitting	overfitting	NOUN
cana-2550	138	7	and	and	CCONJ
cana-2550	138	8	improves	improve	VERB
cana-2550	138	9	the	the	DET
cana-2550	138	10	ability	ability	NOUN
cana-2550	138	11	of	of	ADP
cana-2550	138	12	the	the	DET
cana-2550	138	13	model	model	NOUN
cana-2550	138	14	to	to	PART
cana-2550	138	15	accommodate	accommodate	VERB
cana-2550	138	16	new	new	ADJ
cana-2550	138	17	data	datum	NOUN
cana-2550	138	18	.	.	PUNCT
cana-2550	139	1	moreover	moreover	ADV
cana-2550	139	2	,	,	PUNCT
cana-2550	139	3	we	we	PRON
cana-2550	139	4	introduce	introduce	VERB
cana-2550	139	5	an	an	DET
cana-2550	139	6	adaptive	adaptive	ADJ
cana-2550	139	7	stopping	stopping	NOUN
cana-2550	139	8	rule	rule	NOUN
cana-2550	139	9	that	that	PRON
cana-2550	139	10	terminates	terminate	VERB
cana-2550	139	11	optimization	optimization	NOUN
cana-2550	139	12	when	when	SCONJ
cana-2550	139	13	the	the	DET
cana-2550	139	14	loss	loss	NOUN
cana-2550	139	15	function	function	NOUN
cana-2550	139	16	improvement	improvement	NOUN
cana-2550	139	17	between	between	ADP
cana-2550	139	18	iterations	iteration	NOUN
cana-2550	139	19	dips	dip	NOUN
cana-2550	139	20	below	below	ADP
cana-2550	139	21	a	a	DET
cana-2550	139	22	set	set	VERB
cana-2550	139	23	threshold	threshold	NOUN
cana-2550	139	24	.	.	PUNCT
cana-2550	140	1	this	this	DET
cana-2550	140	2	approach	approach	NOUN
cana-2550	140	3	minimizes	minimize	VERB
cana-2550	140	4	computation	computation	NOUN
cana-2550	140	5	time	time	NOUN
cana-2550	140	6	and	and	CCONJ
cana-2550	140	7	eliminates	eliminate	VERB
cana-2550	140	8	redundant	redundant	ADJ
cana-2550	140	9	updates	update	NOUN
cana-2550	140	10	.	.	PUNCT
cana-2550	141	1	the	the	DET
cana-2550	141	2	formula	formula	NOUN
cana-2550	141	3	for	for	ADP
cana-2550	141	4	the	the	DET
cana-2550	141	5	regularized	regularize	VERB
cana-2550	141	6	loss	loss	NOUN
cana-2550	141	7	function	function	NOUN
cana-2550	141	8	is	be	AUX
cana-2550	141	9	as	as	SCONJ
cana-2550	141	10	follows	follow	VERB
cana-2550	141	11	:	:	PUNCT
cana-2550	141	12	ℒreg(θ	ℒreg(θ	X
cana-2550	141	13	)	)	PUNCT
cana-2550	141	14	=	=	SYM
cana-2550	141	15	ℒ(θ	ℒ(θ	X
cana-2550	141	16	)	)	PUNCT
cana-2550	142	1	+	+	CCONJ
cana-2550	142	2	λ|θ|2	λ|θ|2	PROPN
cana-2550	142	3	ℒ(𝜃	ℒ(𝜃	NUM
cana-2550	142	4	)	)	PUNCT
cana-2550	142	5	is	be	AUX
cana-2550	142	6	the	the	DET
cana-2550	142	7	original	original	ADJ
cana-2550	142	8	loss	loss	NOUN
cana-2550	142	9	function	function	NOUN
cana-2550	142	10	,	,	PUNCT
cana-2550	142	11	λ	λ	PROPN
cana-2550	142	12	is	be	AUX
cana-2550	142	13	the	the	DET
cana-2550	142	14	regularization	regularization	NOUN
cana-2550	142	15	parameter	parameter	NOUN
cana-2550	142	16	,	,	PUNCT
cana-2550	142	17	|θ|2is	|θ|2i	VERB
cana-2550	142	18	the	the	DET
cana-2550	142	19	squared	square	VERB
cana-2550	142	20	l2	l2	NOUN
cana-2550	142	21	norm	norm	NOUN
cana-2550	142	22	of	of	ADP
cana-2550	142	23	the	the	DET
cana-2550	142	24	parameter	parameter	NOUN
cana-2550	142	25	vector	vector	NOUN
cana-2550	142	26	which	which	PRON
cana-2550	142	27	serves	serve	VERB
cana-2550	142	28	as	as	ADP
cana-2550	142	29	the	the	DET
cana-2550	142	30	l2	l2	NOUN
cana-2550	142	31	regularization	regularization	NOUN
cana-2550	142	32	term	term	NOUN
cana-2550	142	33	.	.	PUNCT
cana-2550	143	1	figure	figure	VERB
cana-2550	143	2	4	4	NUM
cana-2550	143	3	time	time	NOUN
cana-2550	143	4	comparision	comparision	NOUN
cana-2550	143	5	table	table	NOUN
cana-2550	143	6	3	3	NUM
cana-2550	143	7	:	:	PUNCT
cana-2550	143	8	time	time	NOUN
cana-2550	143	9	complexity	complexity	NOUN
cana-2550	143	10	comparison	comparison	NOUN
cana-2550	143	11	of	of	ADP
cana-2550	143	12	optimization	optimization	NOUN
cana-2550	143	13	methods	method	NOUN
cana-2550	143	14	optimization	optimization	NOUN
cana-2550	143	15	method	method	NOUN
cana-2550	143	16	time	time	NOUN
cana-2550	143	17	per	per	ADP
cana-2550	143	18	epoch	epoch	NOUN
cana-2550	143	19	(	(	PUNCT
cana-2550	143	20	seconds	second	NOUN
cana-2550	143	21	)	)	PUNCT
cana-2550	143	22	total	total	ADJ
cana-2550	143	23	time	time	NOUN
cana-2550	143	24	for	for	ADP
cana-2550	143	25	convergence	convergence	NOUN
cana-2550	143	26	(	(	PUNCT
cana-2550	143	27	hours	hour	NOUN
cana-2550	143	28	)	)	PUNCT
cana-2550	143	29	stochastic	stochastic	ADJ
cana-2550	143	30	gradient	gradient	ADJ
cana-2550	143	31	descent	descent	NOUN
cana-2550	143	32	(	(	PUNCT
cana-2550	143	33	sgd	sgd	PROPN
cana-2550	143	34	)	)	PUNCT
cana-2550	143	35	1.6	1.6	NUM
cana-2550	143	36	4.0	4.0	NUM
cana-2550	143	37	adam	adam	PROPN
cana-2550	143	38	optimizer	optimizer	NOUN
cana-2550	143	39	2.2	2.2	NUM
cana-2550	143	40	5.3	5.3	NUM
cana-2550	143	41	gradient	gradient	ADJ
cana-2550	143	42	descent	descent	NOUN
cana-2550	143	43	1.3	1.3	NUM
cana-2550	143	44	3.8	3.8	NUM
cana-2550	143	45	adagrad	adagrad	ADJ
cana-2550	143	46	2.4	2.4	NUM
cana-2550	143	47	5.5	5.5	NUM
cana-2550	143	48	momentum	momentum	NOUN
cana-2550	143	49	-	-	PUNCT
cana-2550	143	50	based	base	VERB
cana-2550	143	51	sgd	sgd	PROPN
cana-2550	143	52	1.3	1.3	NUM
cana-2550	143	53	3.2	3.2	NUM
cana-2550	143	54	description	description	NOUN
cana-2550	143	55	:	:	PUNCT
cana-2550	143	56	this	this	DET
cana-2550	143	57	table	table	NOUN
cana-2550	143	58	shows	show	VERB
cana-2550	143	59	the	the	DET
cana-2550	143	60	time	time	NOUN
cana-2550	143	61	complexity	complexity	NOUN
cana-2550	143	62	for	for	ADP
cana-2550	143	63	each	each	DET
cana-2550	143	64	optimization	optimization	NOUN
cana-2550	143	65	method	method	NOUN
cana-2550	143	66	,	,	PUNCT
cana-2550	143	67	highlighting	highlight	VERB
cana-2550	143	68	the	the	DET
cana-2550	143	69	computational	computational	ADJ
cana-2550	143	70	efficiency	efficiency	NOUN
cana-2550	143	71	of	of	ADP
cana-2550	143	72	different	different	ADJ
cana-2550	143	73	approaches	approach	NOUN
cana-2550	143	74	.	.	PUNCT
cana-2550	144	1	while	while	SCONJ
cana-2550	144	2	more	more	ADV
cana-2550	144	3	sophisticated	sophisticated	ADJ
cana-2550	144	4	methods	method	NOUN
cana-2550	144	5	like	like	ADP
cana-2550	144	6	adam	adam	PROPN
cana-2550	144	7	may	may	AUX
cana-2550	144	8	provide	provide	VERB
cana-2550	144	9	better	well	ADJ
cana-2550	144	10	performance	performance	NOUN
cana-2550	144	11	,	,	PUNCT
cana-2550	144	12	they	they	PRON
cana-2550	144	13	tend	tend	VERB
cana-2550	144	14	to	to	PART
cana-2550	144	15	have	have	VERB
cana-2550	144	16	higher	high	ADJ
cana-2550	144	17	time	time	NOUN
cana-2550	144	18	complexity	complexity	NOUN
cana-2550	144	19	compared	compare	VERB
cana-2550	144	20	to	to	ADP
cana-2550	144	21	simpler	simple	ADJ
cana-2550	144	22	methods	method	NOUN
cana-2550	144	23	like	like	ADP
cana-2550	144	24	gradient	gradient	ADJ
cana-2550	144	25	descent	descent	NOUN
cana-2550	144	26	.	.	PUNCT
cana-2550	145	1	communications	communication	NOUN
cana-2550	145	2	on	on	ADP
cana-2550	145	3	applied	apply	VERB
cana-2550	145	4	nonlinear	nonlinear	ADJ
cana-2550	145	5	analysis	analysis	NOUN
cana-2550	145	6	issn	issn	NOUN
cana-2550	145	7	:	:	PUNCT
cana-2550	145	8	1074	1074	NUM
cana-2550	145	9	-	-	PUNCT
cana-2550	145	10	133x	133x	NUM
cana-2550	145	11	vol	vol	NOUN
cana-2550	145	12	32	32	NUM
cana-2550	145	13	no	no	NOUN
cana-2550	145	14	.	.	PUNCT
cana-2550	146	1	3s	3s	NUM
cana-2550	146	2	(	(	PUNCT
cana-2550	146	3	2025	2025	NUM
cana-2550	146	4	)	)	PUNCT
cana-2550	146	5	76	76	NUM
cana-2550	146	6	https://internationalpubls.com	https://internationalpubls.com	X
cana-2550	147	1	6.conclusion	6.conclusion	NUM
cana-2550	147	2	:	:	PUNCT
cana-2550	147	3	in	in	ADP
cana-2550	147	4	this	this	DET
cana-2550	147	5	paper	paper	NOUN
cana-2550	147	6	we	we	PRON
cana-2550	147	7	have	have	AUX
cana-2550	147	8	discussed	discuss	VERB
cana-2550	147	9	the	the	DET
cana-2550	147	10	optimization	optimization	NOUN
cana-2550	147	11	limitations	limitation	NOUN
cana-2550	147	12	in	in	ADP
cana-2550	147	13	neural	neural	ADJ
cana-2550	147	14	networks	network	NOUN
cana-2550	147	15	,	,	PUNCT
cana-2550	147	16	focusing	focus	VERB
cana-2550	147	17	on	on	ADP
cana-2550	147	18	the	the	DET
cana-2550	147	19	importance	importance	NOUN
cana-2550	147	20	of	of	ADP
cana-2550	147	21	mathematical	mathematical	ADJ
cana-2550	147	22	techniques	technique	NOUN
cana-2550	147	23	so	so	SCONJ
cana-2550	147	24	that	that	SCONJ
cana-2550	147	25	the	the	DET
cana-2550	147	26	efficiency	efficiency	NOUN
cana-2550	147	27	of	of	ADP
cana-2550	147	28	deep	deep	ADJ
cana-2550	147	29	learning	learning	NOUN
cana-2550	147	30	architectures	architecture	NOUN
cana-2550	147	31	improves	improve	VERB
cana-2550	147	32	.	.	PUNCT
cana-2550	148	1	optimization	optimization	NOUN
cana-2550	148	2	is	be	AUX
cana-2550	148	3	one	one	NUM
cana-2550	148	4	of	of	ADP
cana-2550	148	5	the	the	DET
cana-2550	148	6	important	important	ADJ
cana-2550	148	7	elements	element	NOUN
cana-2550	148	8	in	in	ADP
cana-2550	148	9	neural	neural	ADJ
cana-2550	148	10	network	network	NOUN
cana-2550	148	11	training	training	NOUN
cana-2550	148	12	because	because	SCONJ
cana-2550	148	13	it	it	PRON
cana-2550	148	14	impacts	impact	VERB
cana-2550	148	15	model	model	NOUN
cana-2550	148	16	accuracy	accuracy	NOUN
cana-2550	148	17	,	,	PUNCT
cana-2550	148	18	convergence	convergence	NOUN
cana-2550	148	19	speed	speed	NOUN
cana-2550	148	20	,	,	PUNCT
cana-2550	148	21	and	and	CCONJ
cana-2550	148	22	computational	computational	ADJ
cana-2550	148	23	efficiency	efficiency	NOUN
cana-2550	148	24	.	.	PUNCT
cana-2550	149	1	the	the	DET
cana-2550	149	2	traditional	traditional	ADJ
cana-2550	149	3	methods	method	NOUN
cana-2550	149	4	of	of	ADP
cana-2550	149	5	gradient	gradient	ADJ
cana-2550	149	6	descent	descent	NOUN
cana-2550	149	7	,	,	PUNCT
cana-2550	149	8	stochastic	stochastic	ADJ
cana-2550	149	9	gradient	gradient	ADJ
cana-2550	149	10	descent	descent	NOUN
cana-2550	149	11	,	,	PUNCT
cana-2550	149	12	and	and	CCONJ
cana-2550	149	13	their	their	PRON
cana-2550	149	14	adaptive	adaptive	ADJ
cana-2550	149	15	variants	variant	NOUN
cana-2550	149	16	such	such	ADJ
cana-2550	149	17	as	as	ADP
cana-2550	149	18	adam	adam	PROPN
cana-2550	149	19	and	and	CCONJ
cana-2550	149	20	rmsprop	rmsprop	NOUN
cana-2550	149	21	have	have	AUX
cana-2550	149	22	been	be	AUX
cana-2550	149	23	the	the	DET
cana-2550	149	24	cornerstones	cornerstone	NOUN
cana-2550	149	25	of	of	ADP
cana-2550	149	26	training	train	VERB
cana-2550	149	27	neural	neural	ADJ
cana-2550	149	28	networks	network	NOUN
cana-2550	149	29	.	.	PUNCT
cana-2550	150	1	however	however	ADV
cana-2550	150	2	,	,	PUNCT
cana-2550	150	3	these	these	DET
cana-2550	150	4	techniques	technique	NOUN
cana-2550	150	5	have	have	VERB
cana-2550	150	6	challenges	challenge	NOUN
cana-2550	150	7	of	of	ADP
cana-2550	150	8	facing	face	VERB
cana-2550	150	9	the	the	DET
cana-2550	150	10	problem	problem	NOUN
cana-2550	150	11	of	of	ADP
cana-2550	150	12	vanishing	vanish	VERB
cana-2550	150	13	or	or	CCONJ
cana-2550	150	14	exploding	explode	VERB
cana-2550	150	15	gradients	gradient	NOUN
cana-2550	150	16	,	,	PUNCT
cana-2550	150	17	slow	slow	ADJ
cana-2550	150	18	convergence	convergence	NOUN
cana-2550	150	19	,	,	PUNCT
cana-2550	150	20	and	and	CCONJ
cana-2550	150	21	sensitivity	sensitivity	NOUN
cana-2550	150	22	to	to	ADP
cana-2550	150	23	the	the	DET
cana-2550	150	24	hyperparameters	hyperparameter	NOUN
cana-2550	150	25	.	.	PUNCT
cana-2550	151	1	to	to	PART
cana-2550	151	2	overcome	overcome	VERB
cana-2550	151	3	such	such	ADJ
cana-2550	151	4	shortcomings	shortcoming	NOUN
cana-2550	151	5	,	,	PUNCT
cana-2550	151	6	we	we	PRON
cana-2550	151	7	proposed	propose	VERB
cana-2550	151	8	an	an	DET
cana-2550	151	9	improved	improved	ADJ
cana-2550	151	10	optimization	optimization	NOUN
cana-2550	151	11	method	method	NOUN
cana-2550	151	12	that	that	PRON
cana-2550	151	13	utilizes	utilize	VERB
cana-2550	151	14	the	the	DET
cana-2550	151	15	principle	principle	NOUN
cana-2550	151	16	of	of	ADP
cana-2550	151	17	momentum	momentum	NOUN
cana-2550	151	18	-	-	PUNCT
cana-2550	151	19	based	base	VERB
cana-2550	151	20	techniques	technique	NOUN
cana-2550	151	21	with	with	ADP
cana-2550	151	22	adaptive	adaptive	ADJ
cana-2550	151	23	learning	learning	NOUN
cana-2550	151	24	rates	rate	NOUN
cana-2550	151	25	.	.	PUNCT
cana-2550	152	1	it	it	PRON
cana-2550	152	2	is	be	AUX
cana-2550	152	3	shown	show	VERB
cana-2550	152	4	that	that	SCONJ
cana-2550	152	5	the	the	DET
cana-2550	152	6	developed	develop	VERB
cana-2550	152	7	approach	approach	NOUN
cana-2550	152	8	performs	perform	VERB
cana-2550	152	9	better	well	ADV
cana-2550	152	10	in	in	ADP
cana-2550	152	11	terms	term	NOUN
cana-2550	152	12	of	of	ADP
cana-2550	152	13	convergence	convergence	NOUN
cana-2550	152	14	speed	speed	NOUN
cana-2550	152	15	,	,	PUNCT
cana-2550	152	16	final	final	ADJ
cana-2550	152	17	training	training	NOUN
cana-2550	152	18	and	and	CCONJ
cana-2550	152	19	validation	validation	NOUN
cana-2550	152	20	loss	loss	NOUN
cana-2550	152	21	,	,	PUNCT
cana-2550	152	22	and	and	CCONJ
cana-2550	152	23	overall	overall	ADJ
cana-2550	152	24	model	model	NOUN
cana-2550	152	25	accuracy	accuracy	NOUN
cana-2550	152	26	through	through	ADP
cana-2550	152	27	empirical	empirical	ADJ
cana-2550	152	28	results	result	NOUN
cana-2550	152	29	obtained	obtain	VERB
cana-2550	152	30	on	on	ADP
cana-2550	152	31	various	various	ADJ
cana-2550	152	32	neural	neural	ADJ
cana-2550	152	33	network	network	NOUN
cana-2550	152	34	architectures	architecture	NOUN
cana-2550	152	35	and	and	CCONJ
cana-2550	152	36	datasets	dataset	NOUN
cana-2550	152	37	.	.	PUNCT
cana-2550	153	1	our	our	PRON
cana-2550	153	2	results	result	NOUN
cana-2550	153	3	point	point	VERB
cana-2550	153	4	out	out	ADP
cana-2550	153	5	the	the	DET
cana-2550	153	6	necessity	necessity	NOUN
cana-2550	153	7	of	of	ADP
cana-2550	153	8	incorporating	incorporate	VERB
cana-2550	153	9	optimization	optimization	NOUN
cana-2550	153	10	techniques	technique	NOUN
cana-2550	153	11	that	that	PRON
cana-2550	153	12	are	be	AUX
cana-2550	153	13	application	application	NOUN
cana-2550	153	14	-	-	PUNCT
cana-2550	153	15	specific	specific	ADJ
cana-2550	153	16	and	and	CCONJ
cana-2550	153	17	architectureand	architectureand	NOUN
cana-2550	153	18	dataset	dataset	NOUN
cana-2550	153	19	-	-	PUNCT
cana-2550	153	20	dependent	dependent	ADJ
cana-2550	153	21	,	,	PUNCT
cana-2550	153	22	thereby	thereby	ADV
cana-2550	153	23	enabling	enable	VERB
cana-2550	153	24	better	well	ADJ
cana-2550	153	25	performance	performance	NOUN
cana-2550	153	26	and	and	CCONJ
cana-2550	153	27	generalization	generalization	NOUN
cana-2550	153	28	of	of	ADP
cana-2550	153	29	the	the	DET
cana-2550	153	30	model	model	NOUN
cana-2550	153	31	.	.	PUNCT
cana-2550	154	1	subsequent	subsequent	ADJ
cana-2550	154	2	work	work	NOUN
cana-2550	154	3	will	will	AUX
cana-2550	154	4	extend	extend	VERB
cana-2550	154	5	the	the	DET
cana-2550	154	6	method	method	NOUN
cana-2550	154	7	to	to	ADP
cana-2550	154	8	distributed	distribute	VERB
cana-2550	154	9	learning	learn	VERB
cana-2550	154	10	scenarios	scenario	NOUN
cana-2550	154	11	and	and	CCONJ
cana-2550	154	12	integrate	integrate	VERB
cana-2550	154	13	it	it	PRON
cana-2550	154	14	with	with	ADP
cana-2550	154	15	nas	na	NOUN
cana-2550	154	16	towards	towards	ADP
cana-2550	154	17	automatic	automatic	ADJ
cana-2550	154	18	and	and	CCONJ
cana-2550	154	19	optimized	optimize	VERB
cana-2550	154	20	design	design	NOUN
cana-2550	154	21	for	for	ADP
cana-2550	154	22	scalable	scalable	ADJ
cana-2550	154	23	and	and	CCONJ
cana-2550	154	24	efficient	efficient	ADJ
cana-2550	154	25	deep	deep	ADJ
cana-2550	154	26	learning	learning	NOUN
cana-2550	154	27	solutions	solution	NOUN
cana-2550	154	28	in	in	ADP
cana-2550	154	29	various	various	ADJ
cana-2550	154	30	applications	application	NOUN
cana-2550	154	31	.	.	PUNCT
cana-2550	155	1	references	reference	NOUN
cana-2550	155	2	:	:	PUNCT
cana-2550	156	1	[	[	X
cana-2550	156	2	1	1	NUM
cana-2550	156	3	]	]	X
cana-2550	156	4	kingma	kingma	PROPN
cana-2550	156	5	,	,	PUNCT
cana-2550	156	6	d.	d.	PROPN
cana-2550	156	7	p.	p.	PROPN
cana-2550	156	8	,	,	PUNCT
cana-2550	156	9	&	&	CCONJ
cana-2550	156	10	ba	ba	PROPN
cana-2550	156	11	,	,	PUNCT
cana-2550	156	12	j.	j.	PROPN
cana-2550	156	13	(	(	PUNCT
cana-2550	156	14	2015	2015	NUM
cana-2550	156	15	)	)	PUNCT
cana-2550	156	16	.	.	PUNCT
cana-2550	157	1	adam	adam	PROPN
cana-2550	157	2	:	:	PUNCT
cana-2550	157	3	a	a	DET
cana-2550	157	4	method	method	NOUN
cana-2550	157	5	for	for	ADP
cana-2550	157	6	stochastic	stochastic	ADJ
cana-2550	157	7	optimization	optimization	NOUN
cana-2550	157	8	.	.	PUNCT
cana-2550	158	1	iclr	iclr	NOUN
cana-2550	158	2	.	.	PUNCT
cana-2550	159	1	[	[	X
cana-2550	159	2	2	2	NUM
cana-2550	159	3	]	]	SYM
cana-2550	159	4	lecun	lecun	ADJ
cana-2550	159	5	,	,	PUNCT
cana-2550	159	6	y.	y.	PROPN
cana-2550	159	7	,	,	PUNCT
cana-2550	159	8	bengio	bengio	PROPN
cana-2550	159	9	,	,	PUNCT
cana-2550	159	10	y.	y.	PROPN
cana-2550	159	11	,	,	PUNCT
cana-2550	159	12	&	&	CCONJ
cana-2550	159	13	hinton	hinton	PROPN
cana-2550	159	14	,	,	PUNCT
cana-2550	159	15	g.	g.	PROPN
cana-2550	159	16	(	(	PUNCT
cana-2550	159	17	2015	2015	NUM
cana-2550	159	18	)	)	PUNCT
cana-2550	159	19	.	.	PUNCT
cana-2550	160	1	deep	deep	ADJ
cana-2550	160	2	learning	learning	NOUN
cana-2550	160	3	.	.	PUNCT
cana-2550	161	1	nature	nature	NOUN
cana-2550	161	2	.	.	PUNCT
cana-2550	162	1	[	[	X
cana-2550	162	2	3	3	NUM
cana-2550	162	3	]	]	X
cana-2550	162	4	bottou	bottou	NOUN
cana-2550	162	5	,	,	PUNCT
cana-2550	162	6	l.	l.	PROPN
cana-2550	162	7	(	(	PUNCT
cana-2550	162	8	2010	2010	NUM
cana-2550	162	9	)	)	PUNCT
cana-2550	162	10	.	.	PUNCT
cana-2550	163	1	large	large	ADJ
cana-2550	163	2	-	-	PUNCT
cana-2550	163	3	scale	scale	NOUN
cana-2550	163	4	machine	machine	NOUN
cana-2550	163	5	learning	learn	VERB
cana-2550	163	6	with	with	ADP
cana-2550	163	7	stochastic	stochastic	ADJ
cana-2550	163	8	gradient	gradient	ADJ
cana-2550	163	9	descent	descent	NOUN
cana-2550	163	10	.	.	PUNCT
cana-2550	164	1	compstat	compstat	PROPN
cana-2550	164	2	.	.	PUNCT
cana-2550	165	1	[	[	X
cana-2550	165	2	4	4	NUM
cana-2550	165	3	]	]	X
cana-2550	165	4	rumelhart	rumelhart	NOUN
cana-2550	165	5	,	,	PUNCT
cana-2550	165	6	d.	d.	PROPN
cana-2550	165	7	e.	e.	PROPN
cana-2550	165	8	,	,	PUNCT
cana-2550	165	9	hinton	hinton	PROPN
cana-2550	165	10	,	,	PUNCT
cana-2550	165	11	g.	g.	PROPN
cana-2550	165	12	e.	e.	PROPN
cana-2550	165	13	,	,	PUNCT
cana-2550	165	14	&	&	CCONJ
cana-2550	165	15	williams	williams	PROPN
cana-2550	165	16	,	,	PUNCT
cana-2550	165	17	r.	r.	PROPN
cana-2550	165	18	j.	j.	PROPN
cana-2550	165	19	(	(	PUNCT
cana-2550	165	20	1986	1986	NUM
cana-2550	165	21	)	)	PUNCT
cana-2550	165	22	.	.	PUNCT
cana-2550	166	1	learning	learn	VERB
cana-2550	166	2	representations	representation	NOUN
cana-2550	166	3	by	by	ADP
cana-2550	166	4	back	back	ADV
cana-2550	166	5	-	-	PUNCT
cana-2550	166	6	propagating	propagate	VERB
cana-2550	166	7	errors	error	NOUN
cana-2550	166	8	.	.	PUNCT
cana-2550	167	1	nature	nature	NOUN
cana-2550	167	2	.	.	PUNCT
cana-2550	168	1	[	[	X
cana-2550	168	2	5	5	X
cana-2550	168	3	]	]	PUNCT
cana-2550	168	4	he	he	PRON
cana-2550	168	5	,	,	PUNCT
cana-2550	168	6	k.	k.	PROPN
cana-2550	168	7	,	,	PUNCT
cana-2550	168	8	zhang	zhang	PROPN
cana-2550	168	9	,	,	PUNCT
cana-2550	168	10	x.	x.	PROPN
cana-2550	168	11	,	,	PUNCT
cana-2550	168	12	ren	ren	PROPN
cana-2550	168	13	,	,	PUNCT
cana-2550	168	14	s.	s.	PROPN
cana-2550	168	15	,	,	PUNCT
cana-2550	168	16	&	&	CCONJ
cana-2550	168	17	sun	sun	PROPN
cana-2550	168	18	,	,	PUNCT
cana-2550	168	19	j.	j.	PROPN
cana-2550	168	20	(	(	PUNCT
cana-2550	168	21	2016	2016	NUM
cana-2550	168	22	)	)	PUNCT
cana-2550	168	23	.	.	PUNCT
cana-2550	169	1	deep	deep	ADJ
cana-2550	169	2	residual	residual	ADJ
cana-2550	169	3	learning	learning	NOUN
cana-2550	169	4	for	for	ADP
cana-2550	169	5	image	image	NOUN
cana-2550	169	6	recognition	recognition	NOUN
cana-2550	169	7	.	.	PUNCT
cana-2550	170	1	cvpr	cvpr	NOUN
cana-2550	170	2	.	.	PUNCT
cana-2550	171	1	[	[	X
cana-2550	171	2	6	6	NUM
cana-2550	171	3	]	]	X
cana-2550	171	4	nocedal	nocedal	NOUN
cana-2550	171	5	,	,	PUNCT
cana-2550	171	6	j.	j.	PROPN
cana-2550	171	7	,	,	PUNCT
cana-2550	171	8	&	&	CCONJ
cana-2550	171	9	wright	wright	PROPN
cana-2550	171	10	,	,	PUNCT
cana-2550	171	11	s.	s.	PROPN
cana-2550	171	12	j.	j.	PROPN
cana-2550	171	13	(	(	PUNCT
cana-2550	171	14	2006	2006	NUM
cana-2550	171	15	)	)	PUNCT
cana-2550	171	16	.	.	PUNCT
cana-2550	172	1	numerical	numerical	PROPN
cana-2550	172	2	optimization	optimization	PROPN
cana-2550	172	3	.	.	PUNCT
cana-2550	173	1	springer	springer	NOUN
cana-2550	173	2	.	.	PUNCT
cana-2550	174	1	[	[	X
cana-2550	174	2	7	7	NUM
cana-2550	174	3	]	]	X
cana-2550	174	4	ioffe	ioffe	NOUN
cana-2550	174	5	,	,	PUNCT
cana-2550	174	6	s.	s.	PROPN
cana-2550	174	7	,	,	PUNCT
cana-2550	174	8	&	&	CCONJ
cana-2550	174	9	szegedy	szegedy	PROPN
cana-2550	174	10	,	,	PUNCT
cana-2550	174	11	c.	c.	NOUN
cana-2550	174	12	(	(	PUNCT
cana-2550	174	13	2015	2015	NUM
cana-2550	174	14	)	)	PUNCT
cana-2550	174	15	.	.	PUNCT
cana-2550	175	1	batch	batch	NOUN
cana-2550	175	2	normalization	normalization	NOUN
cana-2550	175	3	:	:	PUNCT
cana-2550	175	4	accelerating	accelerate	VERB
cana-2550	175	5	deep	deep	ADJ
cana-2550	175	6	network	network	NOUN
cana-2550	175	7	training	training	NOUN
cana-2550	175	8	by	by	ADP
cana-2550	175	9	reducing	reduce	VERB
cana-2550	175	10	internal	internal	ADJ
cana-2550	175	11	covariate	covariate	ADJ
cana-2550	175	12	shift	shift	NOUN
cana-2550	175	13	.	.	PUNCT
cana-2550	176	1	icml	icml	NOUN
cana-2550	176	2	.	.	PUNCT
cana-2550	177	1	[	[	X
cana-2550	177	2	8	8	NUM
cana-2550	177	3	]	]	X
cana-2550	177	4	goodfellow	goodfellow	PROPN
cana-2550	177	5	,	,	PUNCT
cana-2550	177	6	i.	i.	PROPN
cana-2550	177	7	,	,	PUNCT
cana-2550	177	8	bengio	bengio	PROPN
cana-2550	177	9	,	,	PUNCT
cana-2550	177	10	y.	y.	PROPN
cana-2550	177	11	,	,	PUNCT
cana-2550	177	12	&	&	CCONJ
cana-2550	177	13	courville	courville	PROPN
cana-2550	177	14	,	,	PUNCT
cana-2550	177	15	a.	a.	NOUN
cana-2550	177	16	(	(	PUNCT
cana-2550	177	17	2016	2016	NUM
cana-2550	177	18	)	)	PUNCT
cana-2550	177	19	.	.	PUNCT
cana-2550	178	1	deep	deep	ADJ
cana-2550	178	2	learning	learning	NOUN
cana-2550	178	3	.	.	PUNCT
cana-2550	179	1	mit	mit	PROPN
cana-2550	179	2	press	press	NOUN
cana-2550	179	3	.	.	PUNCT
cana-2550	180	1	[	[	X
cana-2550	180	2	9	9	NUM
cana-2550	180	3	]	]	X
cana-2550	180	4	hinton	hinton	PROPN
cana-2550	180	5	,	,	PUNCT
cana-2550	180	6	g.	g.	PROPN
cana-2550	180	7	,	,	PUNCT
cana-2550	180	8	srivastava	srivastava	PROPN
cana-2550	180	9	,	,	PUNCT
cana-2550	180	10	n.	n.	NOUN
cana-2550	180	11	,	,	PUNCT
cana-2550	180	12	&	&	CCONJ
cana-2550	180	13	swersky	swersky	PROPN
cana-2550	180	14	,	,	PUNCT
cana-2550	180	15	k.	k.	PROPN
cana-2550	180	16	(	(	PUNCT
cana-2550	180	17	2012	2012	NUM
cana-2550	180	18	)	)	PUNCT
cana-2550	180	19	.	.	PUNCT
cana-2550	181	1	neural	neural	ADJ
cana-2550	181	2	networks	network	NOUN
cana-2550	181	3	for	for	ADP
cana-2550	181	4	machine	machine	NOUN
cana-2550	181	5	learning	learning	NOUN
cana-2550	181	6	.	.	PUNCT
cana-2550	182	1	lecture	lecture	NOUN
cana-2550	182	2	notes	note	NOUN
cana-2550	182	3	.	.	PUNCT
cana-2550	183	1	[	[	X
cana-2550	183	2	10	10	NUM
cana-2550	183	3	]	]	PUNCT
cana-2550	183	4	sutskever	sutskever	PROPN
cana-2550	183	5	,	,	PUNCT
cana-2550	183	6	i.	i.	PROPN
cana-2550	183	7	,	,	PUNCT
cana-2550	183	8	martens	martens	PROPN
cana-2550	183	9	,	,	PUNCT
cana-2550	183	10	j.	j.	PROPN
cana-2550	183	11	,	,	PUNCT
cana-2550	183	12	dahl	dahl	PROPN
cana-2550	183	13	,	,	PUNCT
cana-2550	183	14	g.	g.	PROPN
cana-2550	183	15	,	,	PUNCT
cana-2550	183	16	&	&	CCONJ
cana-2550	183	17	hinton	hinton	PROPN
cana-2550	183	18	,	,	PUNCT
cana-2550	183	19	g.	g.	PROPN
cana-2550	183	20	(	(	PUNCT
cana-2550	183	21	2013	2013	NUM
cana-2550	183	22	)	)	PUNCT
cana-2550	183	23	.	.	PUNCT
cana-2550	184	1	on	on	ADP
cana-2550	184	2	the	the	DET
cana-2550	184	3	importance	importance	NOUN
cana-2550	184	4	of	of	ADP
cana-2550	184	5	initialization	initialization	NOUN
cana-2550	184	6	and	and	CCONJ
cana-2550	184	7	momentum	momentum	NOUN
cana-2550	184	8	in	in	ADP
cana-2550	184	9	deep	deep	ADJ
cana-2550	184	10	learning	learning	NOUN
cana-2550	184	11	.	.	PUNCT
cana-2550	185	1	icml	icml	NOUN
cana-2550	185	2	.	.	PUNCT
cana-2550	186	1	[	[	X
cana-2550	186	2	11	11	NUM
cana-2550	186	3	]	]	PUNCT
cana-2550	186	4	zeiler	zeiler	NOUN
cana-2550	186	5	,	,	PUNCT
cana-2550	186	6	m.	m.	PROPN
cana-2550	186	7	d.	d.	PROPN
cana-2550	186	8	(	(	PUNCT
cana-2550	186	9	2012	2012	NUM
cana-2550	186	10	)	)	PUNCT
cana-2550	186	11	.	.	PUNCT
cana-2550	187	1	adadelta	adadelta	ADJ
cana-2550	187	2	:	:	PUNCT
cana-2550	187	3	an	an	DET
cana-2550	187	4	adaptive	adaptive	ADJ
cana-2550	187	5	learning	learning	NOUN
cana-2550	187	6	rate	rate	NOUN
cana-2550	187	7	method	method	NOUN
cana-2550	187	8	.	.	PUNCT
cana-2550	188	1	arxiv	arxiv	PROPN
cana-2550	188	2	preprint	preprint	PROPN
cana-2550	188	3	arxiv:1212.5701	arxiv:1212.5701	PROPN
cana-2550	188	4	.	.	PUNCT
cana-2550	189	1	[	[	X
cana-2550	189	2	12	12	NUM
cana-2550	189	3	]	]	X
cana-2550	189	4	y.	y.	NOUN
cana-2550	189	5	(	(	PUNCT
cana-2550	189	6	2012	2012	NUM
cana-2550	189	7	)	)	PUNCT
cana-2550	189	8	.	.	PUNCT
cana-2550	190	1	practical	practical	ADJ
cana-2550	190	2	recommendations	recommendation	NOUN
cana-2550	190	3	for	for	ADP
cana-2550	190	4	gradient	gradient	NOUN
cana-2550	190	5	-	-	PUNCT
cana-2550	190	6	based	base	VERB
cana-2550	190	7	training	training	NOUN
cana-2550	190	8	of	of	ADP
cana-2550	190	9	deep	deep	ADJ
cana-2550	190	10	architectures	architecture	NOUN
cana-2550	190	11	.	.	PUNCT
cana-2550	191	1	neural	neural	ADJ
cana-2550	191	2	networks	network	NOUN
cana-2550	191	3	:	:	PUNCT
cana-2550	191	4	tricks	trick	NOUN
cana-2550	191	5	of	of	ADP
cana-2550	191	6	the	the	DET
cana-2550	191	7	trade	trade	NOUN
cana-2550	191	8	.	.	PUNCT
cana-2550	192	1	[	[	X
cana-2550	192	2	13	13	NUM
cana-2550	192	3	]	]	SYM
cana-2550	192	4	chollet	chollet	NOUN
cana-2550	192	5	,	,	PUNCT
cana-2550	192	6	f.	f.	PROPN
cana-2550	192	7	(	(	PUNCT
cana-2550	192	8	2017	2017	NUM
cana-2550	192	9	)	)	PUNCT
cana-2550	192	10	.	.	PUNCT
cana-2550	193	1	deep	deep	ADJ
cana-2550	193	2	learning	learn	VERB
cana-2550	193	3	with	with	ADP
cana-2550	193	4	python	python	PROPN
cana-2550	193	5	.	.	PUNCT
cana-2550	194	1	manning	man	VERB
cana-2550	194	2	.	.	PUNCT
cana-2550	195	1	[	[	X
cana-2550	195	2	14	14	NUM
cana-2550	195	3	]	]	X
cana-2550	195	4	dean	dean	PROPN
cana-2550	195	5	,	,	PUNCT
cana-2550	195	6	j.	j.	PROPN
cana-2550	195	7	,	,	PUNCT
cana-2550	195	8	et	et	PROPN
cana-2550	195	9	al	al	PROPN
cana-2550	195	10	.	.	PUNCT
cana-2550	195	11	(	(	PUNCT
cana-2550	195	12	2012	2012	NUM
cana-2550	195	13	)	)	PUNCT
cana-2550	195	14	.	.	PUNCT
cana-2550	196	1	large	large	ADJ
cana-2550	196	2	-	-	PUNCT
cana-2550	196	3	scale	scale	NOUN
cana-2550	196	4	distributed	distribute	VERB
cana-2550	196	5	deep	deep	ADJ
cana-2550	196	6	networks	network	NOUN
cana-2550	196	7	.	.	PUNCT
cana-2550	197	1	nips	nip	NOUN
cana-2550	197	2	.	.	PUNCT
cana-2550	198	1	[	[	X
cana-2550	198	2	15	15	NUM
cana-2550	198	3	]	]	X
cana-2550	198	4	reddi	reddi	NOUN
cana-2550	198	5	,	,	PUNCT
cana-2550	198	6	s.	s.	PROPN
cana-2550	198	7	j.	j.	PROPN
cana-2550	198	8	,	,	PUNCT
cana-2550	198	9	kale	kale	PROPN
cana-2550	198	10	,	,	PUNCT
cana-2550	198	11	s.	s.	PROPN
cana-2550	198	12	,	,	PUNCT
cana-2550	198	13	&	&	CCONJ
cana-2550	198	14	kumar	kumar	PROPN
cana-2550	198	15	,	,	PUNCT
cana-2550	198	16	s.	s.	PROPN
cana-2550	198	17	(	(	PUNCT
cana-2550	198	18	2018	2018	NUM
cana-2550	198	19	)	)	PUNCT
cana-2550	198	20	.	.	PUNCT
cana-2550	199	1	on	on	ADP
cana-2550	199	2	the	the	DET
cana-2550	199	3	convergence	convergence	NOUN
cana-2550	199	4	of	of	ADP
cana-2550	199	5	adam	adam	PROPN
cana-2550	199	6	and	and	CCONJ
cana-2550	199	7	beyond	beyond	ADP
cana-2550	199	8	.	.	PUNCT
cana-2550	200	1	iclr	iclr	NOUN
cana-2550	200	2	.	.	PUNCT
cana-2550	201	1	[	[	X
cana-2550	201	2	16	16	NUM
cana-2550	201	3	]	]	PUNCT
cana-2550	201	4	pascanu	pascanu	NOUN
cana-2550	201	5	,	,	PUNCT
cana-2550	201	6	r.	r.	PROPN
cana-2550	201	7	,	,	PUNCT
cana-2550	201	8	mikolov	mikolov	NOUN
cana-2550	201	9	,	,	PUNCT
cana-2550	201	10	t.	t.	PROPN
cana-2550	201	11	,	,	PUNCT
cana-2550	201	12	&	&	CCONJ
cana-2550	201	13	bengio	bengio	PROPN
cana-2550	201	14	,	,	PUNCT
cana-2550	201	15	y.	y.	PROPN
cana-2550	201	16	(	(	PUNCT
cana-2550	201	17	2013	2013	NUM
cana-2550	201	18	)	)	PUNCT
cana-2550	201	19	.	.	PUNCT
cana-2550	202	1	on	on	ADP
cana-2550	202	2	the	the	DET
cana-2550	202	3	difficulty	difficulty	NOUN
cana-2550	202	4	of	of	ADP
cana-2550	202	5	training	training	NOUN
cana-2550	202	6	recurrent	recurrent	ADJ
cana-2550	202	7	neural	neural	ADJ
cana-2550	202	8	networks	network	NOUN
cana-2550	202	9	.	.	PUNCT
cana-2550	203	1	icml	icml	NOUN
cana-2550	203	2	.	.	PUNCT
cana-2550	204	1	[	[	X
cana-2550	204	2	17	17	NUM
cana-2550	204	3	]	]	X
cana-2550	204	4	bishop	bishop	PROPN
cana-2550	204	5	,	,	PUNCT
cana-2550	204	6	c.	c.	PROPN
cana-2550	204	7	m.	m.	NOUN
cana-2550	204	8	(	(	PUNCT
cana-2550	204	9	2006	2006	NUM
cana-2550	204	10	)	)	PUNCT
cana-2550	204	11	.	.	PUNCT
cana-2550	205	1	pattern	pattern	NOUN
cana-2550	205	2	recognition	recognition	NOUN
cana-2550	205	3	and	and	CCONJ
cana-2550	205	4	machine	machine	NOUN
cana-2550	205	5	learning	learning	NOUN
cana-2550	205	6	.	.	PUNCT
cana-2550	206	1	springer	springer	NOUN
cana-2550	206	2	.	.	PUNCT
cana-2550	207	1	[	[	X
cana-2550	207	2	18	18	NUM
cana-2550	207	3	]	]	X
cana-2550	207	4	zhang	zhang	PROPN
cana-2550	207	5	,	,	PUNCT
cana-2550	207	6	y.	y.	PROPN
cana-2550	207	7	,	,	PUNCT
cana-2550	207	8	&	&	CCONJ
cana-2550	207	9	lecun	lecun	PROPN
cana-2550	207	10	,	,	PUNCT
cana-2550	207	11	y.	y.	PROPN
cana-2550	207	12	(	(	PUNCT
cana-2550	207	13	2015	2015	NUM
cana-2550	207	14	)	)	PUNCT
cana-2550	207	15	.	.	PUNCT
cana-2550	208	1	which	which	PRON
cana-2550	208	2	optimization	optimization	NOUN
cana-2550	208	3	algorithm	algorithm	NOUN
cana-2550	208	4	works	work	VERB
cana-2550	208	5	best	good	ADJ
cana-2550	208	6	for	for	ADP
cana-2550	208	7	large	large	ADJ
cana-2550	208	8	-	-	PUNCT
cana-2550	208	9	scale	scale	NOUN
cana-2550	208	10	deep	deep	ADJ
cana-2550	208	11	learning	learning	NOUN
cana-2550	208	12	?	?	PUNCT
cana-2550	209	1	icml	icml	NOUN
cana-2550	209	2	.	.	PUNCT
cana-2550	210	1	[	[	X
cana-2550	210	2	19	19	NUM
cana-2550	210	3	]	]	X
cana-2550	210	4	silver	silver	NOUN
cana-2550	210	5	,	,	PUNCT
cana-2550	210	6	d.	d.	PROPN
cana-2550	210	7	,	,	PUNCT
cana-2550	210	8	et	et	PROPN
cana-2550	210	9	al	al	PROPN
cana-2550	210	10	.	.	PUNCT
cana-2550	210	11	(	(	PUNCT
cana-2550	210	12	2016	2016	NUM
cana-2550	210	13	)	)	PUNCT
cana-2550	210	14	.	.	PUNCT
cana-2550	211	1	mastering	master	VERB
cana-2550	211	2	the	the	DET
cana-2550	211	3	game	game	NOUN
cana-2550	211	4	of	of	ADP
cana-2550	211	5	go	go	NOUN
cana-2550	211	6	with	with	ADP
cana-2550	211	7	deep	deep	ADJ
cana-2550	211	8	neural	neural	ADJ
cana-2550	211	9	networks	network	NOUN
cana-2550	211	10	and	and	CCONJ
cana-2550	211	11	tree	tree	NOUN
cana-2550	211	12	search	search	NOUN
cana-2550	211	13	.	.	PUNCT
cana-2550	212	1	nature	nature	NOUN
cana-2550	212	2	.	.	PUNCT
cana-2550	213	1	[	[	X
cana-2550	213	2	20	20	NUM
cana-2550	213	3	]	]	X
cana-2550	213	4	szegedy	szegedy	PROPN
cana-2550	213	5	,	,	PUNCT
cana-2550	213	6	c.	c.	PROPN
cana-2550	213	7	,	,	PUNCT
cana-2550	213	8	et	et	PROPN
cana-2550	213	9	al	al	PROPN
cana-2550	213	10	.	.	PUNCT
cana-2550	213	11	(	(	PUNCT
cana-2550	213	12	2015	2015	NUM
cana-2550	213	13	)	)	PUNCT
cana-2550	213	14	.	.	PUNCT
cana-2550	214	1	going	go	VERB
cana-2550	214	2	deeper	deeply	ADV
cana-2550	214	3	with	with	ADP
cana-2550	214	4	convolutions	convolution	NOUN
cana-2550	214	5	.	.	PUNCT
cana-2550	215	1	cvpr	cvpr	NOUN
cana-2550	215	2	.	.	PUNCT
cana-2550	216	1	[	[	X
cana-2550	216	2	21	21	NUM
cana-2550	216	3	]	]	X
cana-2550	216	4	liu	liu	PROPN
cana-2550	216	5	,	,	PUNCT
cana-2550	216	6	l.	l.	PROPN
cana-2550	216	7	,	,	PUNCT
cana-2550	216	8	et	et	PROPN
cana-2550	216	9	al	al	PROPN
cana-2550	216	10	.	.	PROPN
cana-2550	216	11	(	(	PUNCT
cana-2550	216	12	2020	2020	NUM
cana-2550	216	13	)	)	PUNCT
cana-2550	216	14	.	.	PUNCT
cana-2550	217	1	on	on	ADP
cana-2550	217	2	the	the	DET
cana-2550	217	3	variance	variance	NOUN
cana-2550	217	4	of	of	ADP
cana-2550	217	5	the	the	DET
cana-2550	217	6	adaptive	adaptive	ADJ
cana-2550	217	7	learning	learning	NOUN
cana-2550	217	8	rate	rate	NOUN
cana-2550	217	9	and	and	CCONJ
cana-2550	217	10	beyond	beyond	ADP
cana-2550	217	11	.	.	PUNCT
cana-2550	218	1	iclr	iclr	NOUN
cana-2550	218	2	.	.	PUNCT
cana-2550	219	1	[	[	X
cana-2550	219	2	22	22	NUM
cana-2550	219	3	]	]	SYM
cana-2550	219	4	li	li	PROPN
cana-2550	219	5	,	,	PUNCT
cana-2550	219	6	y.	y.	PROPN
cana-2550	219	7	,	,	PUNCT
cana-2550	219	8	et	et	PROPN
cana-2550	219	9	al	al	PROPN
cana-2550	219	10	.	.	PUNCT
cana-2550	219	11	(	(	PUNCT
cana-2550	219	12	2017	2017	NUM
cana-2550	219	13	)	)	PUNCT
cana-2550	219	14	.	.	PUNCT
cana-2550	220	1	visualizing	visualize	VERB
cana-2550	220	2	the	the	DET
cana-2550	220	3	loss	loss	NOUN
cana-2550	220	4	landscape	landscape	NOUN
cana-2550	220	5	of	of	ADP
cana-2550	220	6	neural	neural	ADJ
cana-2550	220	7	nets	net	NOUN
cana-2550	220	8	.	.	PUNCT
cana-2550	221	1	nips	nip	NOUN
cana-2550	221	2	.	.	PUNCT
cana-2550	222	1	[	[	X
cana-2550	222	2	23	23	NUM
cana-2550	222	3	]	]	X
cana-2550	222	4	gulcehre	gulcehre	PROPN
cana-2550	222	5	,	,	PUNCT
cana-2550	222	6	c.	c.	PROPN
cana-2550	222	7	,	,	PUNCT
cana-2550	222	8	et	et	PROPN
cana-2550	222	9	al	al	PROPN
cana-2550	222	10	.	.	PUNCT
cana-2550	222	11	(	(	PUNCT
cana-2550	222	12	2016	2016	NUM
cana-2550	222	13	)	)	PUNCT
cana-2550	222	14	.	.	PUNCT
cana-2550	223	1	noisy	noisy	ADJ
cana-2550	223	2	activation	activation	NOUN
cana-2550	223	3	functions	function	NOUN
cana-2550	223	4	.	.	PUNCT
cana-2550	224	1	icml	icml	NOUN
cana-2550	224	2	.	.	PUNCT
cana-2550	225	1	[	[	X
cana-2550	225	2	24	24	NUM
cana-2550	225	3	]	]	SYM
cana-2550	225	4	du	du	X
cana-2550	225	5	,	,	PUNCT
cana-2550	225	6	s.	s.	PROPN
cana-2550	225	7	s.	s.	PROPN
cana-2550	225	8	,	,	PUNCT
cana-2550	225	9	et	et	PROPN
cana-2550	225	10	al	al	PROPN
cana-2550	225	11	.	.	PROPN
cana-2550	226	1	(	(	PUNCT
cana-2550	226	2	2019	2019	NUM
cana-2550	226	3	)	)	PUNCT
cana-2550	226	4	.	.	PUNCT
cana-2550	227	1	gradient	gradient	ADJ
cana-2550	227	2	descent	descent	NOUN
cana-2550	227	3	finds	find	VERB
cana-2550	227	4	global	global	ADJ
cana-2550	227	5	minima	minima	NOUN
cana-2550	227	6	of	of	ADP
cana-2550	227	7	deep	deep	ADJ
cana-2550	227	8	neural	neural	ADJ
cana-2550	227	9	networks	network	NOUN
cana-2550	227	10	.	.	PUNCT
cana-2550	228	1	iclr	iclr	NOUN
cana-2550	228	2	.	.	PUNCT
cana-2550	229	1	[	[	X
cana-2550	229	2	25	25	NUM
cana-2550	229	3	]	]	PUNCT
cana-2550	229	4	keskar	keskar	PROPN
cana-2550	229	5	,	,	PUNCT
cana-2550	229	6	n.	n.	PROPN
cana-2550	229	7	s.	s.	PROPN
cana-2550	229	8	,	,	PUNCT
cana-2550	229	9	et	et	PROPN
cana-2550	229	10	al	al	PROPN
cana-2550	229	11	.	.	PUNCT
cana-2550	229	12	(	(	PUNCT
cana-2550	229	13	2017	2017	NUM
cana-2550	229	14	)	)	PUNCT
cana-2550	229	15	.	.	PUNCT
cana-2550	230	1	on	on	ADP
cana-2550	230	2	large	large	ADJ
cana-2550	230	3	-	-	PUNCT
cana-2550	230	4	batch	batch	NOUN
cana-2550	230	5	training	training	NOUN
cana-2550	230	6	for	for	ADP
cana-2550	230	7	deep	deep	ADJ
cana-2550	230	8	learning	learning	NOUN
cana-2550	230	9	:	:	PUNCT
cana-2550	230	10	generalization	generalization	NOUN
cana-2550	230	11	gap	gap	NOUN
cana-2550	230	12	and	and	CCONJ
cana-2550	230	13	sharp	sharp	ADJ
cana-2550	230	14	minima	minima	PROPN
cana-2550	230	15	.	.	PUNCT
cana-2550	231	1	iclr	iclr	PROPN
cana-2550	231	2	.	.	PUNCT