id	sid	tid	token	lemma	pos
ajst-16801	1	1	academic	academic	ADJ
ajst-16801	1	2	journal	journal	NOUN
ajst-16801	1	3	of	of	ADP
ajst-16801	1	4	science	science	NOUN
ajst-16801	1	5	and	and	CCONJ
ajst-16801	1	6	technology	technology	NOUN
ajst-16801	1	7	issn	issn	NOUN
ajst-16801	1	8	:	:	PUNCT
ajst-16801	1	9	2771	2771	NUM
ajst-16801	1	10	-	-	SYM
ajst-16801	1	11	3032	3032	NUM
ajst-16801	1	12	|	|	NOUN
ajst-16801	1	13	vol	vol	NOUN
ajst-16801	1	14	.	.	PROPN
ajst-16801	2	1	9	9	NUM
ajst-16801	2	2	,	,	PUNCT
ajst-16801	2	3	no	no	INTJ
ajst-16801	2	4	.	.	NOUN
ajst-16801	2	5	1	1	NUM
ajst-16801	2	6	,	,	PUNCT
ajst-16801	2	7	2024	2024	NUM
ajst-16801	2	8	235	235	NUM
ajst-16801	2	9	research	research	NOUN
ajst-16801	2	10	on	on	ADP
ajst-16801	2	11	mushroom	mushroom	NOUN
ajst-16801	2	12	image	image	NOUN
ajst-16801	2	13	classification	classification	NOUN
ajst-16801	2	14	algorithm	algorithm	NOUN
ajst-16801	2	15	based	base	VERB
ajst-16801	2	16	on	on	ADP
ajst-16801	2	17	deep	deep	ADJ
ajst-16801	2	18	sparse	sparse	ADJ
ajst-16801	2	19	dictionary	dictionary	ADJ
ajst-16801	2	20	learning	learn	VERB
ajst-16801	2	21	xiyang	xiyang	PROPN
ajst-16801	2	22	guo	guo	PROPN
ajst-16801	2	23	school	school	PROPN
ajst-16801	2	24	of	of	ADP
ajst-16801	2	25	electrical	electrical	ADJ
ajst-16801	2	26	engineering	engineering	NOUN
ajst-16801	2	27	,	,	PUNCT
ajst-16801	2	28	southwest	southwest	PROPN
ajst-16801	2	29	minzu	minzu	PROPN
ajst-16801	2	30	university	university	PROPN
ajst-16801	2	31	,	,	PUNCT
ajst-16801	2	32	chengdu	chengdu	PROPN
ajst-16801	2	33	,	,	PUNCT
ajst-16801	2	34	610041	610041	NUM
ajst-16801	2	35	,	,	PUNCT
ajst-16801	2	36	china	china	PROPN
ajst-16801	2	37	abstract	abstract	NOUN
ajst-16801	2	38	:	:	PUNCT
ajst-16801	2	39	the	the	DET
ajst-16801	2	40	traditional	traditional	ADJ
ajst-16801	2	41	mushroom	mushroom	NOUN
ajst-16801	2	42	feature	feature	NOUN
ajst-16801	2	43	extraction	extraction	NOUN
ajst-16801	2	44	method	method	NOUN
ajst-16801	2	45	has	have	VERB
ajst-16801	2	46	low	low	ADJ
ajst-16801	2	47	classification	classification	NOUN
ajst-16801	2	48	efficiency	efficiency	NOUN
ajst-16801	2	49	and	and	CCONJ
ajst-16801	2	50	unsatisfactory	unsatisfactory	ADJ
ajst-16801	2	51	effect	effect	NOUN
ajst-16801	2	52	.	.	PUNCT
ajst-16801	3	1	dictionary	dictionary	ADJ
ajst-16801	3	2	learning	learning	NOUN
ajst-16801	3	3	is	be	AUX
ajst-16801	3	4	widely	widely	ADV
ajst-16801	3	5	used	use	VERB
ajst-16801	3	6	in	in	ADP
ajst-16801	3	7	image	image	NOUN
ajst-16801	3	8	classification	classification	NOUN
ajst-16801	3	9	.	.	PUNCT
ajst-16801	4	1	however	however	ADV
ajst-16801	4	2	,	,	PUNCT
ajst-16801	4	3	the	the	DET
ajst-16801	4	4	previous	previous	ADJ
ajst-16801	4	5	work	work	NOUN
ajst-16801	4	6	is	be	AUX
ajst-16801	4	7	to	to	PART
ajst-16801	4	8	learn	learn	VERB
ajst-16801	4	9	dictionaries	dictionary	NOUN
ajst-16801	4	10	in	in	ADP
ajst-16801	4	11	the	the	DET
ajst-16801	4	12	original	original	ADJ
ajst-16801	4	13	space	space	NOUN
ajst-16801	4	14	,	,	PUNCT
ajst-16801	4	15	which	which	PRON
ajst-16801	4	16	limits	limit	VERB
ajst-16801	4	17	the	the	DET
ajst-16801	4	18	performance	performance	NOUN
ajst-16801	4	19	of	of	ADP
ajst-16801	4	20	sparse	sparse	ADJ
ajst-16801	4	21	representation	representation	NOUN
ajst-16801	4	22	classification	classification	NOUN
ajst-16801	4	23	.	.	PUNCT
ajst-16801	5	1	in	in	ADP
ajst-16801	5	2	order	order	NOUN
ajst-16801	5	3	to	to	PART
ajst-16801	5	4	solve	solve	VERB
ajst-16801	5	5	the	the	DET
ajst-16801	5	6	problem	problem	NOUN
ajst-16801	5	7	of	of	ADP
ajst-16801	5	8	spatial	spatial	ADJ
ajst-16801	5	9	redundancy	redundancy	NOUN
ajst-16801	5	10	in	in	ADP
ajst-16801	5	11	traditional	traditional	ADJ
ajst-16801	5	12	convolutional	convolutional	ADJ
ajst-16801	5	13	neural	neural	ADJ
ajst-16801	5	14	networks	network	NOUN
ajst-16801	5	15	and	and	CCONJ
ajst-16801	5	16	the	the	DET
ajst-16801	5	17	weak	weak	ADJ
ajst-16801	5	18	performance	performance	NOUN
ajst-16801	5	19	of	of	ADP
ajst-16801	5	20	deep	deep	ADJ
ajst-16801	5	21	learning	learning	NOUN
ajst-16801	5	22	in	in	ADP
ajst-16801	5	23	small	small	ADJ
ajst-16801	5	24	samples	sample	NOUN
ajst-16801	5	25	,	,	PUNCT
ajst-16801	5	26	an	an	DET
ajst-16801	5	27	improved	improved	ADJ
ajst-16801	5	28	dictionary	dictionary	ADJ
ajst-16801	5	29	learning	learning	NOUN
ajst-16801	5	30	algorithm	algorithm	NOUN
ajst-16801	5	31	,	,	PUNCT
ajst-16801	5	32	deep	deep	ADJ
ajst-16801	5	33	sparse	sparse	ADJ
ajst-16801	5	34	dictionary	dictionary	ADJ
ajst-16801	5	35	learning	learning	NOUN
ajst-16801	5	36	(	(	PUNCT
ajst-16801	5	37	dsdl	dsdl	PROPN
ajst-16801	5	38	)	)	PUNCT
ajst-16801	5	39	,	,	PUNCT
ajst-16801	5	40	is	be	AUX
ajst-16801	5	41	proposed	propose	VERB
ajst-16801	5	42	.	.	PUNCT
ajst-16801	6	1	the	the	DET
ajst-16801	6	2	input	input	NOUN
ajst-16801	6	3	to	to	PART
ajst-16801	6	4	dsdl	dsdl	VERB
ajst-16801	6	5	is	be	AUX
ajst-16801	6	6	not	not	PART
ajst-16801	6	7	a	a	DET
ajst-16801	6	8	matrix	matrix	NOUN
ajst-16801	6	9	gathered	gather	VERB
ajst-16801	6	10	from	from	ADP
ajst-16801	6	11	the	the	DET
ajst-16801	6	12	original	original	ADJ
ajst-16801	6	13	grayscale	grayscale	NOUN
ajst-16801	6	14	image	image	NOUN
ajst-16801	6	15	or	or	CCONJ
ajst-16801	6	16	a	a	DET
ajst-16801	6	17	hand	hand	NOUN
ajst-16801	6	18	-	-	PUNCT
ajst-16801	6	19	created	create	VERB
ajst-16801	6	20	feature	feature	NOUN
ajst-16801	6	21	,	,	PUNCT
ajst-16801	6	22	but	but	CCONJ
ajst-16801	6	23	rather	rather	ADV
ajst-16801	6	24	a	a	DET
ajst-16801	6	25	relatively	relatively	ADV
ajst-16801	6	26	deeper	deep	ADJ
ajst-16801	6	27	feature	feature	NOUN
ajst-16801	6	28	extraction	extraction	NOUN
ajst-16801	6	29	via	via	ADP
ajst-16801	6	30	a	a	DET
ajst-16801	6	31	stack	stack	NOUN
ajst-16801	6	32	autoencoder	autoencoder	NOUN
ajst-16801	6	33	.	.	PUNCT
ajst-16801	7	1	then	then	ADV
ajst-16801	7	2	,	,	PUNCT
ajst-16801	7	3	a	a	DET
ajst-16801	7	4	structured	structured	ADJ
ajst-16801	7	5	dictionary	dictionary	NOUN
ajst-16801	7	6	is	be	AUX
ajst-16801	7	7	designed	design	VERB
ajst-16801	7	8	to	to	PART
ajst-16801	7	9	reconstruct	reconstruct	VERB
ajst-16801	7	10	the	the	DET
ajst-16801	7	11	deep	deep	ADJ
ajst-16801	7	12	features	feature	NOUN
ajst-16801	7	13	according	accord	VERB
ajst-16801	7	14	to	to	ADP
ajst-16801	7	15	different	different	ADJ
ajst-16801	7	16	categories	category	NOUN
ajst-16801	7	17	of	of	ADP
ajst-16801	7	18	distinguishing	distinguish	VERB
ajst-16801	7	19	features	feature	NOUN
ajst-16801	7	20	.	.	PUNCT
ajst-16801	8	1	in	in	ADP
ajst-16801	8	2	addition	addition	NOUN
ajst-16801	8	3	,	,	PUNCT
ajst-16801	8	4	it	it	PRON
ajst-16801	8	5	is	be	AUX
ajst-16801	8	6	necessary	necessary	ADJ
ajst-16801	8	7	to	to	PART
ajst-16801	8	8	learn	learn	VERB
ajst-16801	8	9	the	the	DET
ajst-16801	8	10	associated	associate	VERB
ajst-16801	8	11	structured	structure	VERB
ajst-16801	8	12	projection	projection	NOUN
ajst-16801	8	13	sparse	sparse	ADJ
ajst-16801	8	14	dictionary	dictionary	NOUN
ajst-16801	8	15	to	to	PART
ajst-16801	8	16	ensure	ensure	VERB
ajst-16801	8	17	that	that	SCONJ
ajst-16801	8	18	the	the	DET
ajst-16801	8	19	decoder	decoder	NOUN
ajst-16801	8	20	updates	update	VERB
ajst-16801	8	21	in	in	ADP
ajst-16801	8	22	the	the	DET
ajst-16801	8	23	direction	direction	NOUN
ajst-16801	8	24	of	of	ADP
ajst-16801	8	25	the	the	DET
ajst-16801	8	26	deconvolution	deconvolution	NOUN
ajst-16801	8	27	operator	operator	NOUN
ajst-16801	8	28	error	error	NOUN
ajst-16801	8	29	is	be	AUX
ajst-16801	8	30	minimal	minimal	ADJ
ajst-16801	8	31	.	.	PUNCT
ajst-16801	9	1	by	by	ADP
ajst-16801	9	2	utilizing	utilize	VERB
ajst-16801	9	3	sparse	sparse	ADJ
ajst-16801	9	4	dictionary	dictionary	ADJ
ajst-16801	9	5	learning	learning	NOUN
ajst-16801	9	6	loss	loss	NOUN
ajst-16801	9	7	functions	function	NOUN
ajst-16801	9	8	and	and	CCONJ
ajst-16801	9	9	autoencoder	autoencoder	NOUN
ajst-16801	9	10	loss	loss	NOUN
ajst-16801	9	11	functions	function	NOUN
ajst-16801	9	12	,	,	PUNCT
ajst-16801	9	13	dsdl	dsdl	PROPN
ajst-16801	9	14	can	can	AUX
ajst-16801	9	15	simultaneously	simultaneously	ADV
ajst-16801	9	16	learn	learn	VERB
ajst-16801	9	17	deep	deep	ADJ
ajst-16801	9	18	latent	latent	NOUN
ajst-16801	9	19	features	feature	NOUN
ajst-16801	9	20	and	and	CCONJ
ajst-16801	9	21	corresponding	correspond	VERB
ajst-16801	9	22	dictionary	dictionary	ADJ
ajst-16801	9	23	pairs	pair	NOUN
ajst-16801	9	24	.	.	PUNCT
ajst-16801	10	1	in	in	ADP
ajst-16801	10	2	the	the	DET
ajst-16801	10	3	testing	testing	NOUN
ajst-16801	10	4	phase	phase	NOUN
ajst-16801	10	5	of	of	ADP
ajst-16801	10	6	dsdl	dsdl	PROPN
ajst-16801	10	7	,	,	PUNCT
ajst-16801	10	8	the	the	DET
ajst-16801	10	9	minimum	minimum	ADJ
ajst-16801	10	10	errors	error	NOUN
ajst-16801	10	11	of	of	ADP
ajst-16801	10	12	deep	deep	ADJ
ajst-16801	10	13	feature	feature	NOUN
ajst-16801	10	14	and	and	CCONJ
ajst-16801	10	15	structured	structured	ADJ
ajst-16801	10	16	projection	projection	NOUN
ajst-16801	10	17	components	component	NOUN
ajst-16801	10	18	for	for	ADP
ajst-16801	10	19	different	different	ADJ
ajst-16801	10	20	classes	class	NOUN
ajst-16801	10	21	can	can	AUX
ajst-16801	10	22	be	be	AUX
ajst-16801	10	23	directly	directly	ADV
ajst-16801	10	24	represented	represent	VERB
ajst-16801	10	25	by	by	ADP
ajst-16801	10	26	basic	basic	ADJ
ajst-16801	10	27	matrix	matrix	NOUN
ajst-16801	10	28	multiplication	multiplication	NOUN
ajst-16801	10	29	operations	operation	NOUN
ajst-16801	10	30	.	.	PUNCT
ajst-16801	11	1	experimental	experimental	ADJ
ajst-16801	11	2	results	result	NOUN
ajst-16801	11	3	show	show	VERB
ajst-16801	11	4	that	that	SCONJ
ajst-16801	11	5	the	the	DET
ajst-16801	11	6	proposed	propose	VERB
ajst-16801	11	7	method	method	NOUN
ajst-16801	11	8	achieves	achieve	VERB
ajst-16801	11	9	a	a	DET
ajst-16801	11	10	good	good	ADJ
ajst-16801	11	11	classification	classification	NOUN
ajst-16801	11	12	effect	effect	NOUN
ajst-16801	11	13	on	on	ADP
ajst-16801	11	14	mushroom	mushroom	NOUN
ajst-16801	11	15	images	image	NOUN
ajst-16801	11	16	,	,	PUNCT
ajst-16801	11	17	which	which	PRON
ajst-16801	11	18	shows	show	VERB
ajst-16801	11	19	the	the	DET
ajst-16801	11	20	effectiveness	effectiveness	NOUN
ajst-16801	11	21	of	of	ADP
ajst-16801	11	22	the	the	DET
ajst-16801	11	23	method	method	NOUN
ajst-16801	11	24	.	.	PUNCT
ajst-16801	12	1	keywords	keyword	NOUN
ajst-16801	12	2	:	:	PUNCT
ajst-16801	12	3	dictionary	dictionary	ADJ
ajst-16801	12	4	learning	learning	NOUN
ajst-16801	12	5	;	;	PUNCT
ajst-16801	12	6	sparse	sparse	ADJ
ajst-16801	12	7	representation	representation	NOUN
ajst-16801	12	8	;	;	PUNCT
ajst-16801	12	9	autoencoder	autoencoder	NOUN
ajst-16801	12	10	;	;	PUNCT
ajst-16801	12	11	feature	feature	NOUN
ajst-16801	12	12	extraction	extraction	NOUN
ajst-16801	12	13	.	.	PUNCT
ajst-16801	13	1	1	1	X
ajst-16801	13	2	.	.	X
ajst-16801	13	3	introduction	introduction	NOUN
ajst-16801	13	4	mushroom	mushroom	NOUN
ajst-16801	13	5	poisoning	poisoning	NOUN
ajst-16801	13	6	incidents	incident	NOUN
ajst-16801	13	7	emerge	emerge	VERB
ajst-16801	13	8	in	in	ADP
ajst-16801	13	9	society	society	NOUN
ajst-16801	13	10	,	,	PUNCT
ajst-16801	13	11	people	people	NOUN
ajst-16801	13	12	pay	pay	VERB
ajst-16801	13	13	more	more	ADJ
ajst-16801	13	14	and	and	CCONJ
ajst-16801	13	15	more	more	ADJ
ajst-16801	13	16	attention	attention	NOUN
ajst-16801	13	17	to	to	ADP
ajst-16801	13	18	mushroom	mushroom	NOUN
ajst-16801	13	19	food	food	NOUN
ajst-16801	13	20	safety	safety	NOUN
ajst-16801	13	21	,	,	PUNCT
ajst-16801	13	22	so	so	CCONJ
ajst-16801	13	23	the	the	DET
ajst-16801	13	24	accurate	accurate	ADJ
ajst-16801	13	25	identification	identification	NOUN
ajst-16801	13	26	and	and	CCONJ
ajst-16801	13	27	classification	classification	NOUN
ajst-16801	13	28	of	of	ADP
ajst-16801	13	29	mushrooms	mushroom	NOUN
ajst-16801	13	30	is	be	AUX
ajst-16801	13	31	very	very	ADV
ajst-16801	13	32	important	important	ADJ
ajst-16801	13	33	.	.	PUNCT
ajst-16801	14	1	artificial	artificial	ADJ
ajst-16801	14	2	recognition	recognition	NOUN
ajst-16801	14	3	is	be	AUX
ajst-16801	14	4	limited	limit	VERB
ajst-16801	14	5	by	by	ADP
ajst-16801	14	6	experience	experience	NOUN
ajst-16801	14	7	,	,	PUNCT
ajst-16801	14	8	many	many	ADJ
ajst-16801	14	9	mushrooms	mushroom	NOUN
ajst-16801	14	10	are	be	AUX
ajst-16801	14	11	very	very	ADV
ajst-16801	14	12	similar	similar	ADJ
ajst-16801	14	13	in	in	ADP
ajst-16801	14	14	appearance	appearance	NOUN
ajst-16801	14	15	,	,	PUNCT
ajst-16801	14	16	and	and	CCONJ
ajst-16801	14	17	automatic	automatic	ADJ
ajst-16801	14	18	recognition	recognition	NOUN
ajst-16801	14	19	of	of	ADP
ajst-16801	14	20	mushroom	mushroom	NOUN
ajst-16801	14	21	species	specie	NOUN
ajst-16801	14	22	with	with	ADP
ajst-16801	14	23	the	the	DET
ajst-16801	14	24	help	help	NOUN
ajst-16801	14	25	of	of	ADP
ajst-16801	14	26	computer	computer	NOUN
ajst-16801	14	27	image	image	NOUN
ajst-16801	14	28	recognition	recognition	NOUN
ajst-16801	14	29	technology	technology	NOUN
ajst-16801	14	30	has	have	AUX
ajst-16801	14	31	become	become	VERB
ajst-16801	14	32	a	a	DET
ajst-16801	14	33	research	research	NOUN
ajst-16801	14	34	hotspot	hotspot	NOUN
ajst-16801	14	35	.	.	PUNCT
ajst-16801	15	1	at	at	ADP
ajst-16801	15	2	present	present	ADJ
ajst-16801	15	3	,	,	PUNCT
ajst-16801	15	4	most	most	ADJ
ajst-16801	15	5	of	of	ADP
ajst-16801	15	6	the	the	DET
ajst-16801	15	7	problems	problem	NOUN
ajst-16801	15	8	of	of	ADP
ajst-16801	15	9	mushroom	mushroom	NOUN
ajst-16801	15	10	image	image	NOUN
ajst-16801	15	11	classification	classification	NOUN
ajst-16801	15	12	are	be	AUX
ajst-16801	15	13	solved	solve	VERB
ajst-16801	15	14	by	by	ADP
ajst-16801	15	15	deep	deep	ADJ
ajst-16801	15	16	learning	learning	NOUN
ajst-16801	15	17	methods	method	NOUN
ajst-16801	15	18	,	,	PUNCT
ajst-16801	15	19	such	such	ADJ
ajst-16801	15	20	as	as	ADP
ajst-16801	15	21	neural	neural	ADJ
ajst-16801	15	22	networks	network	NOUN
ajst-16801	15	23	,	,	PUNCT
ajst-16801	15	24	autoencoders	autoencoder	NOUN
ajst-16801	15	25	,	,	PUNCT
ajst-16801	15	26	convolutional	convolutional	ADJ
ajst-16801	15	27	neural	neural	ADJ
ajst-16801	15	28	networks	network	NOUN
ajst-16801	15	29	,	,	PUNCT
ajst-16801	15	30	etc	etc	X
ajst-16801	15	31	.	.	X
ajst-16801	15	32	convolutional	convolutional	ADJ
ajst-16801	15	33	neural	neural	ADJ
ajst-16801	15	34	network	network	NOUN
ajst-16801	15	35	(	(	PUNCT
ajst-16801	15	36	cnn	cnn	PROPN
ajst-16801	15	37	)	)	PUNCT
ajst-16801	15	38	has	have	AUX
ajst-16801	15	39	achieved	achieve	VERB
ajst-16801	15	40	great	great	ADJ
ajst-16801	15	41	success	success	NOUN
ajst-16801	15	42	in	in	ADP
ajst-16801	15	43	computer	computer	NOUN
ajst-16801	15	44	image	image	NOUN
ajst-16801	15	45	recognition	recognition	NOUN
ajst-16801	15	46	,	,	PUNCT
ajst-16801	15	47	and	and	CCONJ
ajst-16801	15	48	its	its	PRON
ajst-16801	15	49	recognition	recognition	NOUN
ajst-16801	15	50	efficiency	efficiency	NOUN
ajst-16801	15	51	is	be	AUX
ajst-16801	15	52	constantly	constantly	ADV
ajst-16801	15	53	improving	improve	VERB
ajst-16801	15	54	.	.	PUNCT
ajst-16801	16	1	however	however	ADV
ajst-16801	16	2	,	,	PUNCT
ajst-16801	16	3	the	the	DET
ajst-16801	16	4	feature	feature	NOUN
ajst-16801	16	5	maps	map	NOUN
ajst-16801	16	6	generated	generate	VERB
ajst-16801	16	7	by	by	ADP
ajst-16801	16	8	traditional	traditional	ADJ
ajst-16801	16	9	convolutional	convolutional	ADJ
ajst-16801	16	10	neural	neural	ADJ
ajst-16801	16	11	networks	network	NOUN
ajst-16801	16	12	still	still	ADV
ajst-16801	16	13	have	have	VERB
ajst-16801	16	14	a	a	DET
ajst-16801	16	15	large	large	ADJ
ajst-16801	16	16	amount	amount	NOUN
ajst-16801	16	17	of	of	ADP
ajst-16801	16	18	redundancy	redundancy	NOUN
ajst-16801	16	19	in	in	ADP
ajst-16801	16	20	spatial	spatial	ADJ
ajst-16801	16	21	dimension	dimension	NOUN
ajst-16801	16	22	,	,	PUNCT
ajst-16801	16	23	and	and	CCONJ
ajst-16801	16	24	the	the	DET
ajst-16801	16	25	processing	processing	NOUN
ajst-16801	16	26	efficiency	efficiency	NOUN
ajst-16801	16	27	is	be	AUX
ajst-16801	16	28	not	not	PART
ajst-16801	16	29	efficient	efficient	ADJ
ajst-16801	16	30	[	[	X
ajst-16801	16	31	1	1	NUM
ajst-16801	16	32	]	]	PUNCT
ajst-16801	16	33	.	.	PUNCT
ajst-16801	17	1	as	as	ADP
ajst-16801	17	2	an	an	DET
ajst-16801	17	3	important	important	ADJ
ajst-16801	17	4	machine	machine	NOUN
ajst-16801	17	5	learning	learning	NOUN
ajst-16801	17	6	model	model	NOUN
ajst-16801	17	7	,	,	PUNCT
ajst-16801	17	8	dictionary	dictionary	ADJ
ajst-16801	17	9	learning	learning	NOUN
ajst-16801	17	10	aims	aim	VERB
ajst-16801	17	11	at	at	ADP
ajst-16801	17	12	sparse	sparse	ADJ
ajst-16801	17	13	representation	representation	NOUN
ajst-16801	17	14	of	of	ADP
ajst-16801	17	15	the	the	DET
ajst-16801	17	16	original	original	ADJ
ajst-16801	17	17	signal	signal	NOUN
ajst-16801	17	18	,	,	PUNCT
ajst-16801	17	19	and	and	CCONJ
ajst-16801	17	20	the	the	DET
ajst-16801	17	21	resulting	result	VERB
ajst-16801	17	22	sparse	sparse	ADJ
ajst-16801	17	23	representation	representation	NOUN
ajst-16801	17	24	is	be	AUX
ajst-16801	17	25	very	very	ADV
ajst-16801	17	26	important	important	ADJ
ajst-16801	17	27	to	to	ADP
ajst-16801	17	28	the	the	DET
ajst-16801	17	29	original	original	ADJ
ajst-16801	17	30	data	datum	NOUN
ajst-16801	17	31	strong	strong	ADJ
ajst-16801	17	32	ability	ability	NOUN
ajst-16801	17	33	of	of	ADP
ajst-16801	17	34	representation	representation	NOUN
ajst-16801	17	35	[	[	X
ajst-16801	17	36	2	2	NUM
ajst-16801	17	37	]	]	PUNCT
ajst-16801	17	38	.	.	PUNCT
ajst-16801	18	1	dictionary	dictionary	ADJ
ajst-16801	18	2	learning	learning	NOUN
ajst-16801	18	3	can	can	AUX
ajst-16801	18	4	compress	compress	VERB
ajst-16801	18	5	most	most	ADJ
ajst-16801	18	6	of	of	ADP
ajst-16801	18	7	the	the	DET
ajst-16801	18	8	redundant	redundant	ADJ
ajst-16801	18	9	information	information	NOUN
ajst-16801	18	10	in	in	ADP
ajst-16801	18	11	the	the	DET
ajst-16801	18	12	original	original	ADJ
ajst-16801	18	13	data	datum	NOUN
ajst-16801	18	14	,	,	PUNCT
ajst-16801	18	15	and	and	CCONJ
ajst-16801	18	16	a	a	DET
ajst-16801	18	17	suitable	suitable	ADJ
ajst-16801	18	18	data	data	NOUN
ajst-16801	18	19	dictionary	dictionary	NOUN
ajst-16801	18	20	can	can	AUX
ajst-16801	18	21	effectively	effectively	ADV
ajst-16801	18	22	reduce	reduce	VERB
ajst-16801	18	23	the	the	DET
ajst-16801	18	24	redundancy	redundancy	NOUN
ajst-16801	18	25	of	of	ADP
ajst-16801	18	26	data	datum	NOUN
ajst-16801	18	27	representation	representation	NOUN
ajst-16801	18	28	and	and	CCONJ
ajst-16801	18	29	improve	improve	VERB
ajst-16801	18	30	the	the	DET
ajst-16801	18	31	discrimination	discrimination	NOUN
ajst-16801	18	32	of	of	ADP
ajst-16801	18	33	data	datum	NOUN
ajst-16801	18	34	,	,	PUNCT
ajst-16801	18	35	so	so	SCONJ
ajst-16801	18	36	as	as	SCONJ
ajst-16801	18	37	to	to	PART
ajst-16801	18	38	obtain	obtain	VERB
ajst-16801	18	39	information	information	NOUN
ajst-16801	18	40	with	with	ADP
ajst-16801	18	41	certain	certain	ADJ
ajst-16801	18	42	use	use	NOUN
ajst-16801	18	43	value	value	NOUN
ajst-16801	18	44	.	.	PUNCT
ajst-16801	19	1	by	by	ADP
ajst-16801	19	2	combining	combine	VERB
ajst-16801	19	3	a	a	DET
ajst-16801	19	4	dictionary	dictionary	ADJ
ajst-16801	19	5	learning	learning	NOUN
ajst-16801	19	6	model	model	NOUN
ajst-16801	19	7	with	with	ADP
ajst-16801	19	8	a	a	DET
ajst-16801	19	9	deep	deep	ADJ
ajst-16801	19	10	neural	neural	ADJ
ajst-16801	19	11	network	network	NOUN
ajst-16801	19	12	,	,	PUNCT
ajst-16801	19	13	the	the	DET
ajst-16801	19	14	deep	deep	ADJ
ajst-16801	19	15	underlying	underlying	ADJ
ajst-16801	19	16	representation	representation	NOUN
ajst-16801	19	17	of	of	ADP
ajst-16801	19	18	the	the	DET
ajst-16801	19	19	original	original	ADJ
ajst-16801	19	20	image	image	NOUN
ajst-16801	19	21	data	datum	NOUN
ajst-16801	19	22	becomes	become	VERB
ajst-16801	19	23	an	an	DET
ajst-16801	19	24	input	input	NOUN
ajst-16801	19	25	to	to	ADP
ajst-16801	19	26	the	the	DET
ajst-16801	19	27	dictionary	dictionary	ADJ
ajst-16801	19	28	learning	learning	PROPN
ajst-16801	19	29	model	model	NOUN
ajst-16801	19	30	,	,	PUNCT
ajst-16801	19	31	contributing	contribute	VERB
ajst-16801	19	32	to	to	ADP
ajst-16801	19	33	more	more	ADV
ajst-16801	19	34	robust	robust	ADJ
ajst-16801	19	35	and	and	CCONJ
ajst-16801	19	36	powerful	powerful	ADJ
ajst-16801	19	37	feature	feature	NOUN
ajst-16801	19	38	descriptions	description	NOUN
ajst-16801	19	39	.	.	PUNCT
ajst-16801	20	1	in	in	ADP
ajst-16801	20	2	2015	2015	NUM
ajst-16801	20	3	,	,	PUNCT
ajst-16801	20	4	shen.l	shen.l	NUM
ajst-16801	20	5	et	et	PROPN
ajst-16801	20	6	al	al	PROPN
ajst-16801	20	7	.	.	PROPN
ajst-16801	20	8	proposed	propose	VERB
ajst-16801	20	9	a	a	DET
ajst-16801	20	10	multilevel	multilevel	ADJ
ajst-16801	20	11	discriminant	discriminant	NOUN
ajst-16801	20	12	dictionary	dictionary	ADJ
ajst-16801	20	13	learning	learn	VERB
ajst-16801	20	14	algorithm	algorithm	NOUN
ajst-16801	20	15	(	(	PUNCT
ajst-16801	20	16	mlddl)[3	mlddl)[3	NUM
ajst-16801	20	17	]	]	PUNCT
ajst-16801	20	18	,	,	PUNCT
ajst-16801	20	19	and	and	CCONJ
ajst-16801	20	20	achieved	achieve	VERB
ajst-16801	20	21	good	good	ADJ
ajst-16801	20	22	results	result	NOUN
ajst-16801	20	23	in	in	ADP
ajst-16801	20	24	large	large	ADJ
ajst-16801	20	25	-	-	PUNCT
ajst-16801	20	26	scale	scale	NOUN
ajst-16801	20	27	image	image	NOUN
ajst-16801	20	28	classification	classification	NOUN
ajst-16801	20	29	.	.	PUNCT
ajst-16801	21	1	in	in	ADP
ajst-16801	21	2	a	a	DET
ajst-16801	21	3	paper	paper	NOUN
ajst-16801	21	4	published	publish	VERB
ajst-16801	21	5	in	in	ADP
ajst-16801	21	6	2016	2016	NUM
ajst-16801	21	7	,	,	PUNCT
ajst-16801	21	8	scholar	scholar	NOUN
ajst-16801	21	9	s.bahrampour	s.bahrampour	NUM
ajst-16801	21	10	et	et	PROPN
ajst-16801	21	11	al	al	PROPN
ajst-16801	21	12	.	.	PROPN
ajst-16801	21	13	proposed	propose	VERB
ajst-16801	21	14	a	a	DET
ajst-16801	21	15	dictionary	dictionary	ADJ
ajst-16801	21	16	learning	learn	VERB
ajst-16801	21	17	algorithm	algorithm	NOUN
ajst-16801	21	18	based	base	VERB
ajst-16801	21	19	on	on	ADP
ajst-16801	21	20	multimode	multimode	PROPN
ajst-16801	21	21	task	task	NOUN
ajst-16801	21	22	-	-	PUNCT
ajst-16801	21	23	driven	drive	VERB
ajst-16801	22	1	[	[	X
ajst-16801	22	2	4	4	NUM
ajst-16801	22	3	]	]	PUNCT
ajst-16801	22	4	,	,	PUNCT
ajst-16801	22	5	which	which	PRON
ajst-16801	22	6	is	be	AUX
ajst-16801	22	7	used	use	VERB
ajst-16801	22	8	in	in	ADP
ajst-16801	22	9	multiview	multiview	NOUN
ajst-16801	22	10	face	face	NOUN
ajst-16801	22	11	recognition	recognition	NOUN
ajst-16801	22	12	,	,	PUNCT
ajst-16801	22	13	multi	multi	ADJ
ajst-16801	22	14	-	-	ADJ
ajst-16801	22	15	view	view	ADJ
ajst-16801	22	16	action	action	NOUN
ajst-16801	22	17	recognition	recognition	NOUN
ajst-16801	22	18	and	and	CCONJ
ajst-16801	22	19	other	other	ADJ
ajst-16801	22	20	fields	field	NOUN
ajst-16801	22	21	.	.	PUNCT
ajst-16801	23	1	in	in	ADP
ajst-16801	23	2	the	the	DET
ajst-16801	23	3	same	same	ADJ
ajst-16801	23	4	year	year	NOUN
ajst-16801	23	5	,	,	PUNCT
ajst-16801	23	6	k.wang	k.wang	PROPN
ajst-16801	23	7	et	et	PROPN
ajst-16801	23	8	al	al	PROPN
ajst-16801	23	9	.	.	PROPN
ajst-16801	23	10	proposed	propose	VERB
ajst-16801	23	11	a	a	DET
ajst-16801	23	12	dictionary	dictionary	ADJ
ajst-16801	23	13	learning	learning	NOUN
ajst-16801	23	14	method	method	NOUN
ajst-16801	23	15	driven	drive	VERB
ajst-16801	23	16	by	by	ADP
ajst-16801	23	17	convolutional	convolutional	ADJ
ajst-16801	23	18	neural	neural	ADJ
ajst-16801	23	19	networks	network	NOUN
ajst-16801	23	20	to	to	PART
ajst-16801	23	21	detect	detect	VERB
ajst-16801	23	22	images	image	NOUN
ajst-16801	23	23	[	[	X
ajst-16801	23	24	5	5	NUM
ajst-16801	23	25	]	]	PUNCT
ajst-16801	23	26	,	,	PUNCT
ajst-16801	23	27	and	and	CCONJ
ajst-16801	23	28	enhanced	enhance	VERB
ajst-16801	23	29	the	the	DET
ajst-16801	23	30	dictionary	dictionary	NOUN
ajst-16801	23	31	's	's	PART
ajst-16801	23	32	ability	ability	NOUN
ajst-16801	23	33	to	to	PART
ajst-16801	23	34	recognize	recognize	VERB
ajst-16801	23	35	discriminant	discriminant	ADJ
ajst-16801	23	36	images	image	NOUN
ajst-16801	23	37	mainly	mainly	ADV
ajst-16801	23	38	by	by	ADP
ajst-16801	23	39	using	use	VERB
ajst-16801	23	40	the	the	DET
ajst-16801	23	41	depth	depth	NOUN
ajst-16801	23	42	information	information	NOUN
ajst-16801	23	43	of	of	ADP
ajst-16801	23	44	feature	feature	NOUN
ajst-16801	23	45	mapping	mapping	NOUN
ajst-16801	23	46	.	.	PUNCT
ajst-16801	24	1	zhang.h	zhang.h	X
ajst-16801	25	1	et	et	PROPN
ajst-16801	25	2	al	al	PROPN
ajst-16801	25	3	.	.	PROPN
ajst-16801	25	4	proposed	propose	VERB
ajst-16801	25	5	a	a	DET
ajst-16801	25	6	nonlinear	nonlinear	ADJ
ajst-16801	25	7	dictionary	dictionary	ADJ
ajst-16801	25	8	learning	learning	NOUN
ajst-16801	25	9	algorithm	algorithm	NOUN
ajst-16801	25	10	based	base	VERB
ajst-16801	25	11	on	on	ADP
ajst-16801	25	12	deep	deep	ADJ
ajst-16801	25	13	neural	neural	ADJ
ajst-16801	25	14	networks	network	NOUN
ajst-16801	26	1	[	[	X
ajst-16801	26	2	6	6	NUM
ajst-16801	26	3	]	]	PUNCT
ajst-16801	26	4	,	,	PUNCT
ajst-16801	26	5	and	and	CCONJ
ajst-16801	26	6	discussed	discuss	VERB
ajst-16801	26	7	the	the	DET
ajst-16801	26	8	improvement	improvement	NOUN
ajst-16801	26	9	of	of	ADP
ajst-16801	26	10	deep	deep	ADJ
ajst-16801	26	11	neural	neural	ADJ
ajst-16801	26	12	network	network	NOUN
ajst-16801	26	13	features	feature	VERB
ajst-16801	26	14	on	on	ADP
ajst-16801	26	15	dictionary	dictionary	ADJ
ajst-16801	26	16	learning	learning	NOUN
ajst-16801	26	17	performance	performance	NOUN
ajst-16801	26	18	.	.	PUNCT
ajst-16801	27	1	s.	s.	PROPN
ajst-16801	27	2	ariyal	ariyal	PROPN
ajst-16801	27	3	et	et	PROPN
ajst-16801	27	4	al	al	PROPN
ajst-16801	27	5	.	.	PUNCT
ajst-16801	28	1	[	[	X
ajst-16801	28	2	7	7	X
ajst-16801	28	3	]	]	PUNCT
ajst-16801	28	4	proposed	propose	VERB
ajst-16801	28	5	the	the	DET
ajst-16801	28	6	deep	deep	ADJ
ajst-16801	28	7	dictionary	dictionary	ADJ
ajst-16801	28	8	learning	learning	PROPN
ajst-16801	28	9	(	(	PUNCT
ajst-16801	28	10	ddl	ddl	PROPN
ajst-16801	28	11	)	)	PUNCT
ajst-16801	28	12	algorithm	algorithm	NOUN
ajst-16801	28	13	in	in	ADP
ajst-16801	28	14	2016	2016	NUM
ajst-16801	28	15	and	and	CCONJ
ajst-16801	28	16	applied	apply	VERB
ajst-16801	28	17	it	it	PRON
ajst-16801	28	18	in	in	ADP
ajst-16801	28	19	tasks	task	NOUN
ajst-16801	28	20	such	such	ADJ
ajst-16801	28	21	as	as	ADP
ajst-16801	28	22	handwriting	handwriting	NOUN
ajst-16801	28	23	font	font	NOUN
ajst-16801	28	24	recognition	recognition	NOUN
ajst-16801	28	25	.	.	PUNCT
ajst-16801	29	1	in	in	ADP
ajst-16801	29	2	2017	2017	NUM
ajst-16801	29	3	,	,	PUNCT
ajst-16801	29	4	m.	m.	NOUN
ajst-16801	29	5	lad	lad	NOUN
ajst-16801	29	6	et	et	PROPN
ajst-16801	29	7	al	al	PROPN
ajst-16801	29	8	proposed	propose	VERB
ajst-16801	29	9	the	the	DET
ajst-16801	29	10	convolutional	convolutional	ADJ
ajst-16801	29	11	dictionary	dictionary	ADJ
ajst-16801	29	12	learning	learn	VERB
ajst-16801	29	13	algorithm	algorithm	NOUN
ajst-16801	30	1	[	[	X
ajst-16801	30	2	8	8	NUM
ajst-16801	30	3	]	]	PUNCT
ajst-16801	30	4	,	,	PUNCT
ajst-16801	30	5	which	which	PRON
ajst-16801	30	6	achieved	achieve	VERB
ajst-16801	30	7	good	good	ADJ
ajst-16801	30	8	results	result	NOUN
ajst-16801	30	9	in	in	ADP
ajst-16801	30	10	image	image	NOUN
ajst-16801	30	11	super	super	ADJ
ajst-16801	30	12	-	-	ADJ
ajst-16801	30	13	resolution	resolution	ADJ
ajst-16801	30	14	reconstruction	reconstruction	NOUN
ajst-16801	30	15	and	and	CCONJ
ajst-16801	30	16	texture	texture	ADJ
ajst-16801	30	17	separation	separation	NOUN
ajst-16801	30	18	.	.	PUNCT
ajst-16801	31	1	v.shinghal	v.shinghal	ADJ
ajst-16801	31	2	applied	apply	VERB
ajst-16801	31	3	robust	robust	ADJ
ajst-16801	31	4	discriminant	discriminant	NOUN
ajst-16801	31	5	deep	deep	ADJ
ajst-16801	31	6	dictionary	dictionary	ADJ
ajst-16801	31	7	learning	learn	VERB
ajst-16801	31	8	algorithm	algorithm	NOUN
ajst-16801	31	9	to	to	ADP
ajst-16801	31	10	hyperspectral	hyperspectral	ADJ
ajst-16801	31	11	image	image	NOUN
ajst-16801	31	12	classification	classification	NOUN
ajst-16801	31	13	in	in	ADP
ajst-16801	31	14	2017	2017	NUM
ajst-16801	31	15	.	.	PUNCT
ajst-16801	32	1	in	in	ADP
ajst-16801	32	2	2017	2017	NUM
ajst-16801	32	3	,	,	PUNCT
ajst-16801	32	4	j.y	j.y	PROPN
ajst-16801	32	5	.	.	PROPN
ajst-16801	32	6	ang	ang	PROPN
ajst-16801	32	7	and	and	CCONJ
ajst-16801	32	8	m.h	m.h	PROPN
ajst-16801	32	9	.	.	PROPN
ajst-16801	32	10	ang	ang	PROPN
ajst-16801	32	11	combined	combine	VERB
ajst-16801	32	12	the	the	DET
ajst-16801	32	13	three	three	NUM
ajst-16801	32	14	models	model	NOUN
ajst-16801	32	15	of	of	ADP
ajst-16801	32	16	attention	attention	NOUN
ajst-16801	32	17	model	model	NOUN
ajst-16801	32	18	,	,	PUNCT
ajst-16801	32	19	crf	crf	PROPN
ajst-16801	32	20	and	and	CCONJ
ajst-16801	32	21	dictionary	dictionary	ADJ
ajst-16801	32	22	learning	learning	NOUN
ajst-16801	32	23	and	and	CCONJ
ajst-16801	32	24	proposed	propose	VERB
ajst-16801	32	25	a	a	DET
ajst-16801	32	26	dictionary	dictionary	ADJ
ajst-16801	32	27	learning	learn	VERB
ajst-16801	32	28	algorithm	algorithm	NOUN
ajst-16801	32	29	for	for	ADP
ajst-16801	32	30	joint	joint	ADJ
ajst-16801	32	31	conditional	conditional	ADJ
ajst-16801	32	32	random	random	ADJ
ajst-16801	32	33	fields	field	NOUN
ajst-16801	32	34	[	[	X
ajst-16801	32	35	9	9	NUM
ajst-16801	32	36	]	]	PUNCT
ajst-16801	32	37	,	,	PUNCT
ajst-16801	32	38	applied	apply	VERB
ajst-16801	32	39	this	this	DET
ajst-16801	32	40	method	method	NOUN
ajst-16801	32	41	to	to	PART
ajst-16801	32	42	image	image	VERB
ajst-16801	32	43	visual	visual	ADJ
ajst-16801	32	44	significance	significance	NOUN
ajst-16801	32	45	analysis	analysis	NOUN
ajst-16801	32	46	,	,	PUNCT
ajst-16801	32	47	and	and	CCONJ
ajst-16801	32	48	used	use	VERB
ajst-16801	32	49	dictionary	dictionary	NOUN
ajst-16801	32	50	learning	learn	VERB
ajst-16801	32	51	for	for	ADP
ajst-16801	32	52	classification	classification	NOUN
ajst-16801	32	53	,	,	PUNCT
ajst-16801	32	54	which	which	PRON
ajst-16801	32	55	could	could	AUX
ajst-16801	32	56	identify	identify	VERB
ajst-16801	32	57	and	and	CCONJ
ajst-16801	32	58	segment	segment	VERB
ajst-16801	32	59	interested	interested	ADJ
ajst-16801	32	60	objects	object	NOUN
ajst-16801	32	61	from	from	ADP
ajst-16801	32	62	images	image	NOUN
ajst-16801	32	63	.	.	PUNCT
ajst-16801	33	1	in	in	ADP
ajst-16801	33	2	2018	2018	NUM
ajst-16801	33	3	,	,	PUNCT
ajst-16801	33	4	v.shinghal	v.shinghal	ADJ
ajst-16801	33	5	et	et	PROPN
ajst-16801	33	6	al	al	PROPN
ajst-16801	33	7	.	.	PROPN
ajst-16801	33	8	proposed	propose	VERB
ajst-16801	33	9	a	a	DET
ajst-16801	33	10	minimization	minimization	NOUN
ajst-16801	33	11	technique	technique	NOUN
ajst-16801	33	12	to	to	PART
ajst-16801	33	13	optimize	optimize	VERB
ajst-16801	33	14	deep	deep	ADJ
ajst-16801	33	15	dictionary	dictionary	ADJ
ajst-16801	33	16	learning	learning	NOUN
ajst-16801	34	1	[	[	X
ajst-16801	34	2	10	10	NUM
ajst-16801	34	3	]	]	PUNCT
ajst-16801	34	4	.	.	PUNCT
ajst-16801	35	1	y.iu	y.iu	PROPN
ajst-16801	35	2	is	be	AUX
ajst-16801	35	3	equivalent	equivalent	ADJ
ajst-16801	35	4	to	to	ADP
ajst-16801	35	5	using	use	VERB
ajst-16801	35	6	dictionary	dictionary	ADJ
ajst-16801	35	7	learning	learning	NOUN
ajst-16801	35	8	to	to	PART
ajst-16801	35	9	motivate	motivate	VERB
ajst-16801	35	10	deep	deep	ADJ
ajst-16801	35	11	networks	network	NOUN
ajst-16801	35	12	for	for	ADP
ajst-16801	35	13	scene	scene	NOUN
ajst-16801	35	14	recognition	recognition	NOUN
ajst-16801	35	15	in	in	ADP
ajst-16801	35	16	2018	2018	NUM
ajst-16801	36	1	[	[	X
ajst-16801	36	2	11	11	NUM
ajst-16801	36	3	]	]	PUNCT
ajst-16801	36	4	.	.	PUNCT
ajst-16801	37	1	in	in	ADP
ajst-16801	37	2	2018	2018	NUM
ajst-16801	37	3	,	,	PUNCT
ajst-16801	37	4	m.lad	m.lad	PROPN
ajst-16801	37	5	proposed	propose	VERB
ajst-16801	37	6	a	a	DET
ajst-16801	37	7	multilayer	multilayer	ADJ
ajst-16801	37	8	sparse	sparse	ADJ
ajst-16801	37	9	convolution	convolution	NOUN
ajst-16801	37	10	dictionary	dictionary	ADJ
ajst-16801	37	11	learning	learn	VERB
ajst-16801	37	12	algorithm	algorithm	NOUN
ajst-16801	37	13	(	(	PUNCT
ajst-16801	37	14	mlcsc)[12	mlcsc)[12	PROPN
ajst-16801	37	15	]	]	PUNCT
ajst-16801	37	16	.	.	PUNCT
ajst-16801	38	1	j.u	j.u	PROPN
ajst-16801	38	2	et	et	PROPN
ajst-16801	38	3	al	al	PROPN
ajst-16801	38	4	.	.	PROPN
ajst-16801	38	5	proposed	propose	VERB
ajst-16801	38	6	a	a	DET
ajst-16801	38	7	nonlinear	nonlinear	ADJ
ajst-16801	38	8	dictionary	dictionary	ADJ
ajst-16801	38	9	learning	learning	NOUN
ajst-16801	38	10	algorithm	algorithm	NOUN
ajst-16801	38	11	(	(	PUNCT
ajst-16801	38	12	ndl	ndl	PROPN
ajst-16801	38	13	)	)	PUNCT
ajst-16801	38	14	and	and	CCONJ
ajst-16801	38	15	a	a	DET
ajst-16801	38	16	supervised	supervised	ADJ
ajst-16801	38	17	nonlinear	nonlinear	ADJ
ajst-16801	38	18	dictionary	dictionary	ADJ
ajst-16801	38	19	learning	learning	NOUN
ajst-16801	38	20	algorithm	algorithm	PROPN
ajst-16801	38	21	(	(	PUNCT
ajst-16801	38	22	sndl	sndl	NOUN
ajst-16801	38	23	)	)	PUNCT
ajst-16801	38	24	based	base	VERB
ajst-16801	38	25	on	on	ADP
ajst-16801	38	26	neural	neural	ADJ
ajst-16801	38	27	networks	network	NOUN
ajst-16801	38	28	[	[	X
ajst-16801	38	29	13	13	NUM
ajst-16801	38	30	]	]	PUNCT
ajst-16801	38	31	,	,	PUNCT
ajst-16801	38	32	and	and	CCONJ
ajst-16801	38	33	sndl	sndl	NOUN
ajst-16801	38	34	is	be	AUX
ajst-16801	38	35	mainly	mainly	ADV
ajst-16801	38	36	applied	apply	VERB
ajst-16801	38	37	to	to	ADP
ajst-16801	38	38	image	image	NOUN
ajst-16801	38	39	classification	classification	NOUN
ajst-16801	38	40	tasks	task	NOUN
ajst-16801	38	41	.	.	PUNCT
ajst-16801	39	1	in	in	ADP
ajst-16801	39	2	2019	2019	NUM
ajst-16801	39	3	,	,	PUNCT
ajst-16801	39	4	tang	tang	X
ajst-16801	39	5	et	et	PROPN
ajst-16801	39	6	al	al	PROPN
ajst-16801	39	7	proposed	propose	VERB
ajst-16801	39	8	deep	deep	ADJ
ajst-16801	39	9	microdictionary	microdictionary	NOUN
ajst-16801	39	10	learning	learn	VERB
ajst-16801	39	11	236	236	NUM
ajst-16801	39	12	coding	code	VERB
ajst-16801	39	13	network	network	NOUN
ajst-16801	39	14	(	(	PUNCT
ajst-16801	39	15	ddlcn)[14	ddlcn)[14	PROPN
ajst-16801	39	16	]	]	X
ajst-16801	39	17	,	,	PUNCT
ajst-16801	39	18	which	which	PRON
ajst-16801	39	19	can	can	AUX
ajst-16801	39	20	be	be	AUX
ajst-16801	39	21	used	use	VERB
ajst-16801	39	22	for	for	ADP
ajst-16801	39	23	image	image	NOUN
ajst-16801	39	24	representation	representation	NOUN
ajst-16801	39	25	and	and	CCONJ
ajst-16801	39	26	classification	classification	NOUN
ajst-16801	39	27	.	.	PUNCT
ajst-16801	40	1	although	although	SCONJ
ajst-16801	40	2	the	the	DET
ajst-16801	40	3	above	above	ADJ
ajst-16801	40	4	networks	network	NOUN
ajst-16801	40	5	have	have	VERB
ajst-16801	40	6	great	great	ADJ
ajst-16801	40	7	innovations	innovation	NOUN
ajst-16801	40	8	in	in	ADP
ajst-16801	40	9	algorithms	algorithm	NOUN
ajst-16801	40	10	,	,	PUNCT
ajst-16801	40	11	they	they	PRON
ajst-16801	40	12	are	be	AUX
ajst-16801	40	13	basically	basically	ADV
ajst-16801	40	14	trained	train	VERB
ajst-16801	40	15	on	on	ADP
ajst-16801	40	16	large	large	ADJ
ajst-16801	40	17	data	datum	NOUN
ajst-16801	40	18	sets	set	NOUN
ajst-16801	40	19	.	.	PUNCT
ajst-16801	41	1	in	in	ADP
ajst-16801	41	2	the	the	DET
ajst-16801	41	3	actual	actual	ADJ
ajst-16801	41	4	mushroom	mushroom	NOUN
ajst-16801	41	5	identification	identification	NOUN
ajst-16801	41	6	,	,	PUNCT
ajst-16801	41	7	small	small	ADJ
ajst-16801	41	8	batch	batch	NOUN
ajst-16801	41	9	data	datum	NOUN
ajst-16801	41	10	need	need	VERB
ajst-16801	41	11	to	to	PART
ajst-16801	41	12	be	be	AUX
ajst-16801	41	13	classified	classify	VERB
ajst-16801	41	14	,	,	PUNCT
ajst-16801	41	15	and	and	CCONJ
ajst-16801	41	16	more	more	ADJ
ajst-16801	41	17	network	network	NOUN
ajst-16801	41	18	parameters	parameter	NOUN
ajst-16801	41	19	will	will	AUX
ajst-16801	41	20	cause	cause	VERB
ajst-16801	41	21	the	the	DET
ajst-16801	41	22	above	above	ADJ
ajst-16801	41	23	network	network	NOUN
ajst-16801	41	24	can	can	AUX
ajst-16801	41	25	not	not	PART
ajst-16801	41	26	be	be	AUX
ajst-16801	41	27	fully	fully	ADV
ajst-16801	41	28	trained	train	VERB
ajst-16801	41	29	or	or	CCONJ
ajst-16801	41	30	produce	produce	VERB
ajst-16801	41	31	overfitting	overfitting	NOUN
ajst-16801	41	32	.	.	PUNCT
ajst-16801	42	1	the	the	DET
ajst-16801	42	2	data	datum	NOUN
ajst-16801	42	3	set	set	VERB
ajst-16801	42	4	used	use	VERB
ajst-16801	42	5	in	in	ADP
ajst-16801	42	6	this	this	DET
ajst-16801	42	7	paper	paper	NOUN
ajst-16801	42	8	is	be	AUX
ajst-16801	42	9	small	small	ADJ
ajst-16801	42	10	.	.	PUNCT
ajst-16801	43	1	in	in	ADP
ajst-16801	43	2	order	order	NOUN
ajst-16801	43	3	to	to	PART
ajst-16801	43	4	reduce	reduce	VERB
ajst-16801	43	5	the	the	DET
ajst-16801	43	6	influence	influence	NOUN
ajst-16801	43	7	caused	cause	VERB
ajst-16801	43	8	by	by	ADP
ajst-16801	43	9	overfitting	overfitte	VERB
ajst-16801	43	10	and	and	CCONJ
ajst-16801	43	11	solve	solve	VERB
ajst-16801	43	12	the	the	DET
ajst-16801	43	13	spatial	spatial	ADJ
ajst-16801	43	14	redundancy	redundancy	NOUN
ajst-16801	43	15	problem	problem	NOUN
ajst-16801	43	16	in	in	ADP
ajst-16801	43	17	traditional	traditional	ADJ
ajst-16801	43	18	convolutional	convolutional	ADJ
ajst-16801	43	19	neural	neural	ADJ
ajst-16801	43	20	networks	network	NOUN
ajst-16801	43	21	,	,	PUNCT
ajst-16801	43	22	a	a	DET
ajst-16801	43	23	mushroom	mushroom	NOUN
ajst-16801	43	24	image	image	NOUN
ajst-16801	43	25	classification	classification	NOUN
ajst-16801	43	26	and	and	CCONJ
ajst-16801	43	27	recognition	recognition	NOUN
ajst-16801	43	28	method	method	NOUN
ajst-16801	43	29	under	under	ADP
ajst-16801	43	30	small	small	ADJ
ajst-16801	43	31	samples	sample	NOUN
ajst-16801	43	32	,	,	PUNCT
ajst-16801	43	33	namely	namely	ADV
ajst-16801	43	34	deep	deep	ADJ
ajst-16801	43	35	sparse	sparse	ADJ
ajst-16801	43	36	dictionary	dictionary	ADJ
ajst-16801	43	37	learning	learning	NOUN
ajst-16801	43	38	(	(	PUNCT
ajst-16801	43	39	dsdl	dsdl	PROPN
ajst-16801	43	40	)	)	PUNCT
ajst-16801	43	41	,	,	PUNCT
ajst-16801	43	42	is	be	AUX
ajst-16801	43	43	proposed	propose	VERB
ajst-16801	43	44	.	.	PUNCT
ajst-16801	44	1	this	this	DET
ajst-16801	44	2	method	method	NOUN
ajst-16801	44	3	extracts	extract	NOUN
ajst-16801	44	4	relatively	relatively	ADV
ajst-16801	44	5	deeper	deep	ADJ
ajst-16801	44	6	features	feature	NOUN
ajst-16801	44	7	through	through	ADP
ajst-16801	44	8	the	the	DET
ajst-16801	44	9	stack	stack	NOUN
ajst-16801	44	10	autoencoder	autoencoder	NOUN
ajst-16801	44	11	,	,	PUNCT
ajst-16801	44	12	and	and	CCONJ
ajst-16801	44	13	then	then	ADV
ajst-16801	44	14	designs	design	VERB
ajst-16801	44	15	a	a	DET
ajst-16801	44	16	structured	structured	ADJ
ajst-16801	44	17	dictionary	dictionary	NOUN
ajst-16801	44	18	according	accord	VERB
ajst-16801	44	19	to	to	ADP
ajst-16801	44	20	different	different	ADJ
ajst-16801	44	21	categories	category	NOUN
ajst-16801	44	22	of	of	ADP
ajst-16801	44	23	discriminant	discriminant	NOUN
ajst-16801	44	24	features	feature	VERB
ajst-16801	44	25	to	to	PART
ajst-16801	44	26	reconstruct	reconstruct	VERB
ajst-16801	44	27	the	the	DET
ajst-16801	44	28	deep	deep	ADJ
ajst-16801	44	29	features	feature	NOUN
ajst-16801	44	30	.	.	PUNCT
ajst-16801	45	1	2	2	X
ajst-16801	45	2	.	.	X
ajst-16801	45	3	fundamental	fundamental	ADJ
ajst-16801	45	4	2.1	2.1	NUM
ajst-16801	45	5	.	.	PUNCT
ajst-16801	45	6	basic	basic	ADJ
ajst-16801	45	7	theory	theory	NOUN
ajst-16801	45	8	of	of	ADP
ajst-16801	45	9	sparse	sparse	ADJ
ajst-16801	45	10	dictionary	dictionary	ADJ
ajst-16801	45	11	learning	learning	NOUN
ajst-16801	45	12	in	in	ADP
ajst-16801	45	13	the	the	DET
ajst-16801	45	14	traditional	traditional	ADJ
ajst-16801	45	15	sparse	sparse	ADJ
ajst-16801	45	16	representation	representation	NOUN
ajst-16801	45	17	classification	classification	NOUN
ajst-16801	45	18	method	method	NOUN
ajst-16801	45	19	,	,	PUNCT
ajst-16801	45	20	the	the	DET
ajst-16801	45	21	original	original	ADJ
ajst-16801	45	22	training	training	NOUN
ajst-16801	45	23	sample	sample	NOUN
ajst-16801	45	24	is	be	AUX
ajst-16801	45	25	generally	generally	ADV
ajst-16801	45	26	used	use	VERB
ajst-16801	45	27	as	as	ADP
ajst-16801	45	28	the	the	DET
ajst-16801	45	29	dictionary	dictionary	NOUN
ajst-16801	45	30	,	,	PUNCT
ajst-16801	45	31	and	and	CCONJ
ajst-16801	45	32	the	the	DET
ajst-16801	45	33	sparse	sparse	ADJ
ajst-16801	45	34	representation	representation	NOUN
ajst-16801	45	35	matrix	matrix	NOUN
ajst-16801	45	36	is	be	AUX
ajst-16801	45	37	approximates	approximate	NOUN
ajst-16801	45	38	to	to	ADP
ajst-16801	45	39	the	the	DET
ajst-16801	45	40	ideal	ideal	ADJ
ajst-16801	45	41	sparse	sparse	ADJ
ajst-16801	45	42	representation	representation	NOUN
ajst-16801	45	43	through	through	ADP
ajst-16801	45	44	various	various	ADJ
ajst-16801	45	45	constraints	constraint	NOUN
ajst-16801	45	46	,	,	PUNCT
ajst-16801	45	47	and	and	CCONJ
ajst-16801	45	48	only	only	ADV
ajst-16801	45	49	the	the	DET
ajst-16801	45	50	sparse	sparse	ADJ
ajst-16801	45	51	coefficient	coefficient	NOUN
ajst-16801	45	52	is	be	AUX
ajst-16801	45	53	required	require	VERB
ajst-16801	45	54	to	to	PART
ajst-16801	45	55	solve	solve	VERB
ajst-16801	45	56	,	,	PUNCT
ajst-16801	45	57	so	so	CCONJ
ajst-16801	45	58	the	the	DET
ajst-16801	45	59	training	training	NOUN
ajst-16801	45	60	and	and	CCONJ
ajst-16801	45	61	solving	solving	NOUN
ajst-16801	45	62	process	process	NOUN
ajst-16801	45	63	is	be	AUX
ajst-16801	45	64	simple	simple	ADJ
ajst-16801	45	65	.	.	PUNCT
ajst-16801	46	1	however	however	ADV
ajst-16801	46	2	,	,	PUNCT
ajst-16801	46	3	due	due	ADP
ajst-16801	46	4	to	to	ADP
ajst-16801	46	5	the	the	DET
ajst-16801	46	6	simple	simple	ADJ
ajst-16801	46	7	structure	structure	NOUN
ajst-16801	46	8	of	of	ADP
ajst-16801	46	9	the	the	DET
ajst-16801	46	10	dictionary	dictionary	NOUN
ajst-16801	46	11	,	,	PUNCT
ajst-16801	46	12	the	the	DET
ajst-16801	46	13	representation	representation	NOUN
ajst-16801	46	14	coefficient	coefficient	NOUN
ajst-16801	46	15	can	can	AUX
ajst-16801	46	16	not	not	PART
ajst-16801	46	17	reach	reach	VERB
ajst-16801	46	18	the	the	DET
ajst-16801	46	19	ideal	ideal	ADJ
ajst-16801	46	20	sparse	sparse	ADJ
ajst-16801	46	21	structure	structure	NOUN
ajst-16801	46	22	,	,	PUNCT
ajst-16801	46	23	which	which	PRON
ajst-16801	46	24	is	be	AUX
ajst-16801	46	25	not	not	PART
ajst-16801	46	26	conducive	conducive	ADJ
ajst-16801	46	27	to	to	ADP
ajst-16801	46	28	the	the	DET
ajst-16801	46	29	classification	classification	NOUN
ajst-16801	46	30	of	of	ADP
ajst-16801	46	31	complex	complex	ADJ
ajst-16801	46	32	images	image	NOUN
ajst-16801	46	33	.	.	PUNCT
ajst-16801	47	1	in	in	ADP
ajst-16801	47	2	order	order	NOUN
ajst-16801	47	3	to	to	PART
ajst-16801	47	4	approach	approach	VERB
ajst-16801	47	5	the	the	DET
ajst-16801	47	6	optimal	optimal	ADJ
ajst-16801	47	7	sparse	sparse	ADJ
ajst-16801	47	8	representation	representation	NOUN
ajst-16801	47	9	and	and	CCONJ
ajst-16801	47	10	obtain	obtain	VERB
ajst-16801	47	11	better	well	ADJ
ajst-16801	47	12	classification	classification	NOUN
ajst-16801	47	13	results	result	NOUN
ajst-16801	47	14	,	,	PUNCT
ajst-16801	47	15	some	some	DET
ajst-16801	47	16	scholars	scholar	NOUN
ajst-16801	47	17	put	put	VERB
ajst-16801	47	18	forward	forward	ADV
ajst-16801	47	19	dictionary	dictionary	ADJ
ajst-16801	47	20	learning	learning	NOUN
ajst-16801	47	21	methods	method	NOUN
ajst-16801	47	22	to	to	PART
ajst-16801	47	23	obtain	obtain	VERB
ajst-16801	47	24	discriminative	discriminative	NOUN
ajst-16801	47	25	dictionaries	dictionary	NOUN
ajst-16801	47	26	.	.	PUNCT
ajst-16801	48	1	compared	compare	VERB
ajst-16801	48	2	with	with	ADP
ajst-16801	48	3	traditional	traditional	ADJ
ajst-16801	48	4	sparse	sparse	ADJ
ajst-16801	48	5	representation	representation	NOUN
ajst-16801	48	6	classification	classification	NOUN
ajst-16801	48	7	methods	method	NOUN
ajst-16801	48	8	,	,	PUNCT
ajst-16801	48	9	dictionary	dictionary	ADJ
ajst-16801	48	10	learning	learning	NOUN
ajst-16801	48	11	methods	method	NOUN
ajst-16801	48	12	not	not	PART
ajst-16801	48	13	only	only	ADV
ajst-16801	48	14	learn	learn	VERB
ajst-16801	48	15	sparse	sparse	ADJ
ajst-16801	48	16	representation	representation	NOUN
ajst-16801	48	17	coefficients	coefficient	NOUN
ajst-16801	48	18	,	,	PUNCT
ajst-16801	48	19	but	but	CCONJ
ajst-16801	48	20	also	also	ADV
ajst-16801	48	21	learn	learn	VERB
ajst-16801	48	22	dictionaries	dictionary	NOUN
ajst-16801	48	23	.	.	PUNCT
ajst-16801	49	1	the	the	DET
ajst-16801	49	2	step	step	NOUN
ajst-16801	49	3	of	of	ADP
ajst-16801	49	4	the	the	DET
ajst-16801	49	5	dictionary	dictionary	ADJ
ajst-16801	49	6	learning	learning	PROPN
ajst-16801	49	7	model	model	NOUN
ajst-16801	49	8	is	be	AUX
ajst-16801	49	9	to	to	PART
ajst-16801	49	10	obtain	obtain	VERB
ajst-16801	49	11	the	the	DET
ajst-16801	49	12	initial	initial	ADJ
ajst-16801	49	13	dictionary	dictionary	NOUN
ajst-16801	49	14	,	,	PUNCT
ajst-16801	49	15	which	which	PRON
ajst-16801	49	16	can	can	AUX
ajst-16801	49	17	be	be	AUX
ajst-16801	49	18	obtained	obtain	VERB
ajst-16801	49	19	by	by	ADP
ajst-16801	49	20	matrix	matrix	NOUN
ajst-16801	49	21	decomposition	decomposition	NOUN
ajst-16801	49	22	,	,	PUNCT
ajst-16801	49	23	or	or	CCONJ
ajst-16801	49	24	the	the	DET
ajst-16801	49	25	training	training	NOUN
ajst-16801	49	26	set	set	NOUN
ajst-16801	49	27	can	can	AUX
ajst-16801	49	28	be	be	AUX
ajst-16801	49	29	used	use	VERB
ajst-16801	49	30	as	as	ADP
ajst-16801	49	31	the	the	DET
ajst-16801	49	32	initial	initial	ADJ
ajst-16801	49	33	dictionary	dictionary	NOUN
ajst-16801	49	34	.	.	PUNCT
ajst-16801	50	1	then	then	ADV
ajst-16801	50	2	the	the	DET
ajst-16801	50	3	sparse	sparse	ADJ
ajst-16801	50	4	coefficient	coefficient	NOUN
ajst-16801	50	5	is	be	AUX
ajst-16801	50	6	solved	solve	VERB
ajst-16801	50	7	,	,	PUNCT
ajst-16801	50	8	and	and	CCONJ
ajst-16801	50	9	then	then	ADV
ajst-16801	50	10	the	the	DET
ajst-16801	50	11	dictionary	dictionary	ADJ
ajst-16801	50	12	and	and	CCONJ
ajst-16801	50	13	sparse	sparse	ADJ
ajst-16801	50	14	coefficient	coefficient	NOUN
ajst-16801	50	15	are	be	AUX
ajst-16801	50	16	updated	update	VERB
ajst-16801	50	17	alternately	alternately	ADV
ajst-16801	50	18	.	.	PUNCT
ajst-16801	51	1	the	the	DET
ajst-16801	51	2	algorithm	algorithm	NOUN
ajst-16801	51	3	flow	flow	NOUN
ajst-16801	51	4	is	be	AUX
ajst-16801	51	5	shown	show	VERB
ajst-16801	51	6	in	in	ADP
ajst-16801	51	7	figure	figure	NOUN
ajst-16801	51	8	1	1	NUM
ajst-16801	51	9	.	.	PUNCT
ajst-16801	51	10	figure	figure	NOUN
ajst-16801	51	11	1	1	NUM
ajst-16801	51	12	.	.	PUNCT
ajst-16801	52	1	the	the	DET
ajst-16801	52	2	flow	flow	NOUN
ajst-16801	52	3	chart	chart	NOUN
ajst-16801	52	4	of	of	ADP
ajst-16801	52	5	sparse	sparse	ADJ
ajst-16801	52	6	dictionary	dictionary	NOUN
ajst-16801	52	7	learning	learn	VERB
ajst-16801	52	8	the	the	DET
ajst-16801	52	9	following	follow	VERB
ajst-16801	52	10	is	be	AUX
ajst-16801	52	11	a	a	DET
ajst-16801	52	12	brief	brief	ADJ
ajst-16801	52	13	introduction	introduction	NOUN
ajst-16801	52	14	of	of	ADP
ajst-16801	52	15	classic	classic	ADJ
ajst-16801	52	16	dictionary	dictionary	ADJ
ajst-16801	52	17	learning	learning	NOUN
ajst-16801	52	18	algorithms	algorithm	NOUN
ajst-16801	52	19	such	such	ADJ
ajst-16801	52	20	as	as	ADP
ajst-16801	52	21	mod[15	mod[15	NOUN
ajst-16801	52	22	]	]	PUNCT
ajst-16801	52	23	,	,	PUNCT
ajst-16801	52	24	odl[16	odl[16	PROPN
ajst-16801	52	25	]	]	PUNCT
ajst-16801	52	26	and	and	CCONJ
ajst-16801	52	27	ksvd[17	ksvd[17	PROPN
ajst-16801	52	28	]	]	X
ajst-16801	52	29	.	.	NOUN
ajst-16801	53	1	1	1	X
ajst-16801	53	2	)	)	PUNCT
ajst-16801	53	3	optimal	optimal	ADJ
ajst-16801	53	4	direction	direction	NOUN
ajst-16801	53	5	dictionary	dictionary	NOUN
ajst-16801	53	6	learning	learn	VERB
ajst-16801	53	7	the	the	DET
ajst-16801	53	8	optimal	optimal	ADJ
ajst-16801	53	9	direction	direction	NOUN
ajst-16801	53	10	algorithm	algorithm	NOUN
ajst-16801	53	11	(	(	PUNCT
ajst-16801	53	12	mod	mod	PROPN
ajst-16801	53	13	)	)	PUNCT
ajst-16801	53	14	was	be	AUX
ajst-16801	53	15	first	first	ADV
ajst-16801	53	16	proposed	propose	VERB
ajst-16801	53	17	by	by	ADP
ajst-16801	53	18	engan	engan	NOUN
ajst-16801	53	19	in	in	ADP
ajst-16801	53	20	1999	1999	NUM
ajst-16801	53	21	.	.	PUNCT
ajst-16801	54	1	it	it	PRON
ajst-16801	54	2	selects	select	VERB
ajst-16801	54	3	a	a	DET
ajst-16801	54	4	direction	direction	NOUN
ajst-16801	54	5	with	with	ADP
ajst-16801	54	6	the	the	DET
ajst-16801	54	7	fastest	fast	ADJ
ajst-16801	54	8	error	error	NOUN
ajst-16801	54	9	reduction	reduction	NOUN
ajst-16801	54	10	while	while	SCONJ
ajst-16801	54	11	using	use	VERB
ajst-16801	54	12	the	the	DET
ajst-16801	54	13	least	least	ADJ
ajst-16801	54	14	square	square	ADJ
ajst-16801	54	15	method	method	NOUN
ajst-16801	54	16	to	to	PART
ajst-16801	54	17	update	update	VERB
ajst-16801	54	18	the	the	DET
ajst-16801	54	19	dictionary	dictionary	ADJ
ajst-16801	54	20	and	and	CCONJ
ajst-16801	54	21	sparse	sparse	ADJ
ajst-16801	54	22	coefficient	coefficient	NOUN
ajst-16801	54	23	alternately	alternately	ADV
ajst-16801	54	24	.	.	PUNCT
ajst-16801	55	1	its	its	PRON
ajst-16801	55	2	objective	objective	ADJ
ajst-16801	55	3	function	function	NOUN
ajst-16801	55	4	is	be	AUX
ajst-16801	55	5	as	as	SCONJ
ajst-16801	55	6	follows	follow	VERB
ajst-16801	55	7	:	:	PUNCT
ajst-16801	55	8	min	min	NOUN
ajst-16801	55	9	∑	∑	PROPN
ajst-16801	55	10	∥	∥	PROPN
ajst-16801	55	11	𝑥	𝑥	NOUN
ajst-16801	55	12	∥	∥	NOUN
ajst-16801	55	13	,	,	PUNCT
ajst-16801	55	14	𝑠.	𝑠.	NOUN
ajst-16801	55	15	𝑡.	𝑡.	NOUN
ajst-16801	55	16	∥	∥	PRON
ajst-16801	55	17	𝑇	𝑇	PROPN
ajst-16801	55	18	𝐷𝑋	𝐷𝑋	PROPN
ajst-16801	55	19	∥	∥	PUNCT
ajst-16801	55	20	𝜀	𝜀	X
ajst-16801	55	21	(	(	PUNCT
ajst-16801	55	22	1	1	NUM
ajst-16801	55	23	)	)	PUNCT
ajst-16801	55	24	where	where	SCONJ
ajst-16801	55	25	t	t	PROPN
ajst-16801	55	26	represents	represent	VERB
ajst-16801	55	27	the	the	DET
ajst-16801	55	28	training	training	NOUN
ajst-16801	55	29	sample	sample	NOUN
ajst-16801	55	30	,	,	PUNCT
ajst-16801	55	31	d	d	PROPN
ajst-16801	55	32	is	be	AUX
ajst-16801	55	33	the	the	DET
ajst-16801	55	34	complete	complete	PROPN
ajst-16801	55	35	dictionary	dictionary	NOUN
ajst-16801	55	36	,	,	PUNCT
ajst-16801	55	37	x	x	X
ajst-16801	55	38	is	be	AUX
ajst-16801	55	39	the	the	DET
ajst-16801	55	40	sparse	sparse	ADJ
ajst-16801	55	41	coefficient,𝑇	coefficient,𝑇	NOUN
ajst-16801	55	42	is	be	AUX
ajst-16801	55	43	the	the	DET
ajst-16801	55	44	sparse	sparse	ADJ
ajst-16801	55	45	threshold,∥	threshold,∥	NOUN
ajst-16801	55	46	∎	∎	PROPN
ajst-16801	55	47	∥	∥	PUNCT
ajst-16801	55	48	is	be	AUX
ajst-16801	55	49	the	the	DET
ajst-16801	55	50	frobenius	frobenius	ADJ
ajst-16801	55	51	norm	norm	NOUN
ajst-16801	55	52	of	of	ADP
ajst-16801	55	53	the	the	DET
ajst-16801	55	54	matrix	matrix	NOUN
ajst-16801	55	55	.	.	PUNCT
ajst-16801	56	1	the	the	DET
ajst-16801	56	2	dictionary	dictionary	ADJ
ajst-16801	56	3	updating	updating	NOUN
ajst-16801	56	4	process	process	NOUN
ajst-16801	56	5	of	of	ADP
ajst-16801	56	6	mod	mod	ADJ
ajst-16801	56	7	algorithm	algorithm	NOUN
ajst-16801	56	8	is	be	AUX
ajst-16801	56	9	simple	simple	ADJ
ajst-16801	56	10	,	,	PUNCT
ajst-16801	56	11	but	but	CCONJ
ajst-16801	56	12	it	it	PRON
ajst-16801	56	13	faces	face	VERB
ajst-16801	56	14	np	np	ADJ
ajst-16801	56	15	-	-	PUNCT
ajst-16801	56	16	hard	hard	ADJ
ajst-16801	56	17	problem	problem	NOUN
ajst-16801	56	18	and	and	CCONJ
ajst-16801	56	19	can	can	AUX
ajst-16801	56	20	not	not	PART
ajst-16801	56	21	obtain	obtain	VERB
ajst-16801	56	22	global	global	ADJ
ajst-16801	56	23	optimal	optimal	ADJ
ajst-16801	56	24	solution	solution	NOUN
ajst-16801	56	25	.	.	PUNCT
ajst-16801	57	1	2	2	X
ajst-16801	57	2	)	)	PUNCT
ajst-16801	57	3	online	online	NOUN
ajst-16801	57	4	dictionary	dictionary	NOUN
ajst-16801	57	5	learning	learn	VERB
ajst-16801	57	6	online	online	PROPN
ajst-16801	57	7	dictionary	dictionary	ADJ
ajst-16801	57	8	learning	learning	PROPN
ajst-16801	57	9	(	(	PUNCT
ajst-16801	57	10	odl	odl	PROPN
ajst-16801	57	11	)	)	PUNCT
ajst-16801	57	12	solves	solve	VERB
ajst-16801	57	13	the	the	DET
ajst-16801	57	14	non	non	ADJ
ajst-16801	57	15	-	-	ADJ
ajst-16801	57	16	convex	convex	ADJ
ajst-16801	57	17	problem	problem	NOUN
ajst-16801	57	18	of	of	ADP
ajst-16801	57	19	mod	mod	ADJ
ajst-16801	57	20	algorithm	algorithm	PROPN
ajst-16801	57	21	,	,	PUNCT
ajst-16801	57	22	that	that	ADV
ajst-16801	57	23	is	is	ADV
ajst-16801	57	24	,	,	PUNCT
ajst-16801	57	25	ℓ	ℓ	PROPN
ajst-16801	57	26	norm	norm	NOUN
ajst-16801	57	27	is	be	AUX
ajst-16801	57	28	used	use	VERB
ajst-16801	57	29	instead	instead	ADV
ajst-16801	57	30	of	of	ADP
ajst-16801	57	31	ℓ	ℓ	PROPN
ajst-16801	57	32	norm	norm	NOUN
ajst-16801	57	33	in	in	ADP
ajst-16801	57	34	mod	mod	PROPN
ajst-16801	57	35	model	model	NOUN
ajst-16801	57	36	,	,	PUNCT
ajst-16801	57	37	and	and	CCONJ
ajst-16801	57	38	its	its	PRON
ajst-16801	57	39	objective	objective	ADJ
ajst-16801	57	40	function	function	NOUN
ajst-16801	57	41	is	be	AUX
ajst-16801	57	42	as	as	SCONJ
ajst-16801	57	43	follows	follow	VERB
ajst-16801	57	44	:	:	PUNCT
ajst-16801	58	1	min	min	NOUN
ajst-16801	58	2	d	d	NOUN
ajst-16801	58	3	,	,	PUNCT
ajst-16801	58	4	x	x	X
ajst-16801	58	5	∥	∥	PUNCT
ajst-16801	58	6	𝑇	𝑇	NOUN
ajst-16801	58	7	𝐷𝑋	𝐷𝑋	PROPN
ajst-16801	58	8	∥	∥	PUNCT
ajst-16801	58	9	𝜆	𝜆	ADP
ajst-16801	58	10	∑	∑	PUNCT
ajst-16801	58	11	∥	∥	PROPN
ajst-16801	58	12	𝑥	𝑥	NOUN
ajst-16801	58	13	∥	∥	X
ajst-16801	58	14	(	(	PUNCT
ajst-16801	58	15	2	2	NUM
ajst-16801	58	16	)	)	PUNCT
ajst-16801	58	17	where	where	SCONJ
ajst-16801	58	18	𝜆	𝜆	NOUN
ajst-16801	58	19	is	be	AUX
ajst-16801	58	20	the	the	DET
ajst-16801	58	21	weight	weight	NOUN
ajst-16801	58	22	parameter	parameter	NOUN
ajst-16801	58	23	of	of	ADP
ajst-16801	58	24	the	the	DET
ajst-16801	58	25	constraint	constraint	NOUN
ajst-16801	58	26	term	term	NOUN
ajst-16801	58	27	of	of	ADP
ajst-16801	58	28	the	the	DET
ajst-16801	58	29	ℓ	ℓ	PROPN
ajst-16801	58	30	norm	norm	NOUN
ajst-16801	58	31	.	.	PUNCT
ajst-16801	59	1	in	in	ADP
ajst-16801	59	2	the	the	DET
ajst-16801	59	3	solution	solution	NOUN
ajst-16801	59	4	,	,	PUNCT
ajst-16801	59	5	the	the	DET
ajst-16801	59	6	dictionary	dictionary	ADJ
ajst-16801	59	7	d	d	PROPN
ajst-16801	59	8	and	and	CCONJ
ajst-16801	59	9	the	the	DET
ajst-16801	59	10	representation	representation	NOUN
ajst-16801	59	11	coefficient	coefficient	NOUN
ajst-16801	59	12	a	a	PRON
ajst-16801	59	13	are	be	AUX
ajst-16801	59	14	updated	update	VERB
ajst-16801	59	15	by	by	ADP
ajst-16801	59	16	alternate	alternate	ADJ
ajst-16801	59	17	iteration	iteration	NOUN
ajst-16801	59	18	.	.	PUNCT
ajst-16801	60	1	3	3	X
ajst-16801	60	2	)	)	PUNCT
ajst-16801	60	3	k	k	ADJ
ajst-16801	60	4	-	-	PUNCT
ajst-16801	60	5	svd	svd	PROPN
ajst-16801	60	6	dictionary	dictionary	NOUN
ajst-16801	60	7	learning	learn	VERB
ajst-16801	60	8	although	although	SCONJ
ajst-16801	60	9	its	its	PRON
ajst-16801	60	10	objective	objective	ADJ
ajst-16801	60	11	function	function	NOUN
ajst-16801	60	12	is	be	AUX
ajst-16801	60	13	the	the	DET
ajst-16801	60	14	same	same	ADJ
ajst-16801	60	15	as	as	ADP
ajst-16801	60	16	the	the	DET
ajst-16801	60	17	mod	mod	PROPN
ajst-16801	60	18	algorithm	algorithm	PROPN
ajst-16801	60	19	,	,	PUNCT
ajst-16801	60	20	the	the	DET
ajst-16801	60	21	dictionary	dictionary	ADJ
ajst-16801	60	22	solving	solving	NOUN
ajst-16801	60	23	process	process	NOUN
ajst-16801	60	24	is	be	AUX
ajst-16801	60	25	different	different	ADJ
ajst-16801	60	26	from	from	ADP
ajst-16801	60	27	the	the	DET
ajst-16801	60	28	mod	mod	PROPN
ajst-16801	60	29	algorithm	algorithm	NOUN
ajst-16801	60	30	.	.	PUNCT
ajst-16801	61	1	k	k	X
ajst-16801	61	2	-	-	PUNCT
ajst-16801	61	3	svd	svd	PROPN
ajst-16801	61	4	adopts	adopt	VERB
ajst-16801	61	5	the	the	DET
ajst-16801	61	6	way	way	NOUN
ajst-16801	61	7	of	of	ADP
ajst-16801	61	8	updating	update	VERB
ajst-16801	61	9	the	the	DET
ajst-16801	61	10	dictionary	dictionary	ADJ
ajst-16801	61	11	column	column	NOUN
ajst-16801	61	12	by	by	ADP
ajst-16801	61	13	column	column	NOUN
ajst-16801	61	14	,	,	PUNCT
ajst-16801	61	15	that	that	ADV
ajst-16801	61	16	is	is	ADV
ajst-16801	61	17	,	,	PUNCT
ajst-16801	61	18	fixing	fix	VERB
ajst-16801	61	19	other	other	ADJ
ajst-16801	61	20	column	column	NOUN
ajst-16801	61	21	dictionary	dictionary	ADJ
ajst-16801	61	22	atoms	atom	NOUN
ajst-16801	61	23	and	and	CCONJ
ajst-16801	61	24	updating	update	VERB
ajst-16801	61	25	only	only	ADV
ajst-16801	61	26	one	one	NUM
ajst-16801	61	27	column	column	NOUN
ajst-16801	61	28	atom	atom	NOUN
ajst-16801	61	29	at	at	ADP
ajst-16801	61	30	the	the	DET
ajst-16801	61	31	same	same	ADJ
ajst-16801	61	32	time	time	NOUN
ajst-16801	61	33	,	,	PUNCT
ajst-16801	61	34	so	so	CCONJ
ajst-16801	61	35	it	it	PRON
ajst-16801	61	36	has	have	VERB
ajst-16801	61	37	the	the	DET
ajst-16801	61	38	characteristics	characteristic	NOUN
ajst-16801	61	39	of	of	ADP
ajst-16801	61	40	fast	fast	ADJ
ajst-16801	61	41	convergence	convergence	NOUN
ajst-16801	61	42	.	.	PUNCT
ajst-16801	62	1	updates	update	NOUN
ajst-16801	62	2	to	to	ADP
ajst-16801	62	3	the	the	DET
ajst-16801	62	4	current	current	ADJ
ajst-16801	62	5	dictionary	dictionary	NOUN
ajst-16801	62	6	and	and	CCONJ
ajst-16801	62	7	presentation	presentation	NOUN
ajst-16801	62	8	coefficients	coefficient	NOUN
ajst-16801	62	9	are	be	AUX
ajst-16801	62	10	also	also	ADV
ajst-16801	62	11	done	do	VERB
ajst-16801	62	12	using	use	VERB
ajst-16801	62	13	alternate	alternate	ADJ
ajst-16801	62	14	iterations	iteration	NOUN
ajst-16801	62	15	.	.	PUNCT
ajst-16801	63	1	k	k	X
ajst-16801	63	2	-	-	PUNCT
ajst-16801	63	3	svd	svd	PROPN
ajst-16801	63	4	has	have	AUX
ajst-16801	63	5	been	be	AUX
ajst-16801	63	6	widely	widely	ADV
ajst-16801	63	7	used	use	VERB
ajst-16801	63	8	in	in	ADP
ajst-16801	63	9	pattern	pattern	NOUN
ajst-16801	63	10	recognition	recognition	NOUN
ajst-16801	63	11	and	and	CCONJ
ajst-16801	63	12	image	image	NOUN
ajst-16801	63	13	classification	classification	NOUN
ajst-16801	63	14	because	because	SCONJ
ajst-16801	63	15	of	of	ADP
ajst-16801	63	16	its	its	PRON
ajst-16801	63	17	fast	fast	ADJ
ajst-16801	63	18	convergence	convergence	NOUN
ajst-16801	63	19	.	.	PUNCT
ajst-16801	64	1	in	in	ADP
ajst-16801	64	2	addition	addition	NOUN
ajst-16801	64	3	,	,	PUNCT
ajst-16801	64	4	with	with	ADP
ajst-16801	64	5	the	the	DET
ajst-16801	64	6	increasing	increase	VERB
ajst-16801	64	7	demand	demand	NOUN
ajst-16801	64	8	for	for	ADP
ajst-16801	64	9	classification	classification	NOUN
ajst-16801	64	10	or	or	CCONJ
ajst-16801	64	11	recognition	recognition	NOUN
ajst-16801	64	12	tasks	task	NOUN
ajst-16801	64	13	,	,	PUNCT
ajst-16801	64	14	researchers	researcher	NOUN
ajst-16801	64	15	have	have	AUX
ajst-16801	64	16	mainly	mainly	ADV
ajst-16801	64	17	studied	study	VERB
ajst-16801	64	18	and	and	CCONJ
ajst-16801	64	19	improved	improved	ADJ
ajst-16801	64	20	dictionaries	dictionary	NOUN
ajst-16801	64	21	and	and	CCONJ
ajst-16801	64	22	sparse	sparse	ADJ
ajst-16801	64	23	representation	representation	NOUN
ajst-16801	64	24	coefficients	coefficient	NOUN
ajst-16801	64	25	,	,	PUNCT
ajst-16801	64	26	and	and	CCONJ
ajst-16801	64	27	many	many	ADJ
ajst-16801	64	28	new	new	ADJ
ajst-16801	64	29	dictionary	dictionary	ADJ
ajst-16801	64	30	learning	learning	NOUN
ajst-16801	64	31	methods	method	NOUN
ajst-16801	64	32	based	base	VERB
ajst-16801	64	33	on	on	ADP
ajst-16801	64	34	the	the	DET
ajst-16801	64	35	above	above	ADJ
ajst-16801	64	36	models	model	NOUN
ajst-16801	64	37	have	have	AUX
ajst-16801	64	38	emerged	emerge	VERB
ajst-16801	64	39	.	.	PUNCT
ajst-16801	65	1	therefore	therefore	ADV
ajst-16801	65	2	,	,	PUNCT
ajst-16801	65	3	the	the	DET
ajst-16801	65	4	traditional	traditional	ADJ
ajst-16801	65	5	single	single	ADJ
ajst-16801	65	6	-	-	PUNCT
ajst-16801	65	7	layer	layer	NOUN
ajst-16801	65	8	dictionary	dictionary	ADJ
ajst-16801	65	9	learning	learning	NOUN
ajst-16801	65	10	method	method	NOUN
ajst-16801	65	11	model	model	NOUN
ajst-16801	65	12	can	can	AUX
ajst-16801	65	13	be	be	AUX
ajst-16801	65	14	summarized	summarize	VERB
ajst-16801	65	15	as	as	SCONJ
ajst-16801	65	16	follows	follow	VERB
ajst-16801	65	17	:	:	PUNCT
ajst-16801	66	1	min	min	NOUN
ajst-16801	66	2	d	d	NOUN
ajst-16801	66	3	,	,	PUNCT
ajst-16801	66	4	x	x	X
ajst-16801	66	5	∥	∥	PUNCT
ajst-16801	66	6	𝑇	𝑇	NOUN
ajst-16801	66	7	𝐷𝑋	𝐷𝑋	PROPN
ajst-16801	66	8	∥	∥	PUNCT
ajst-16801	66	9	𝜆	𝜆	X
ajst-16801	66	10	∥	∥	X
ajst-16801	66	11	𝑋	𝑋	NOUN
ajst-16801	66	12	∥	∥	PUNCT
ajst-16801	66	13	𝜓	𝜓	PROPN
ajst-16801	66	14	𝐷	𝐷	PROPN
ajst-16801	66	15	,	,	PUNCT
ajst-16801	66	16	𝑋	𝑋	PROPN
ajst-16801	66	17	,	,	PUNCT
ajst-16801	66	18	𝑌	𝑌	PROPN
ajst-16801	66	19	(	(	PUNCT
ajst-16801	66	20	3	3	NUM
ajst-16801	66	21	)	)	PUNCT
ajst-16801	66	22	where	where	SCONJ
ajst-16801	66	23	,	,	PUNCT
ajst-16801	66	24	y	y	PROPN
ajst-16801	66	25	is	be	AUX
ajst-16801	66	26	the	the	DET
ajst-16801	66	27	label	label	NOUN
ajst-16801	66	28	matrix	matrix	NOUN
ajst-16801	66	29	,	,	PUNCT
ajst-16801	66	30	p=0,1,2	p=0,1,2	NUM
ajst-16801	66	31	,	,	PUNCT
ajst-16801	66	32	representing	represent	VERB
ajst-16801	66	33	the	the	DET
ajst-16801	66	34	type	type	NOUN
ajst-16801	66	35	of	of	ADP
ajst-16801	66	36	ℓ	ℓ	PROPN
ajst-16801	66	37	norm	norm	NOUN
ajst-16801	66	38	constraint	constraint	NOUN
ajst-16801	66	39	,	,	PUNCT
ajst-16801	66	40	and	and	CCONJ
ajst-16801	66	41	𝜓	𝜓	PROPN
ajst-16801	66	42	(	(	PUNCT
ajst-16801	66	43	d	d	NOUN
ajst-16801	66	44	,	,	PUNCT
ajst-16801	66	45	x	x	NOUN
ajst-16801	66	46	,	,	PUNCT
ajst-16801	66	47	y	y	PROPN
ajst-16801	66	48	)	)	PUNCT
ajst-16801	66	49	is	be	AUX
ajst-16801	66	50	the	the	DET
ajst-16801	66	51	discriminant	discriminant	ADJ
ajst-16801	66	52	term	term	NOUN
ajst-16801	66	53	.	.	PUNCT
ajst-16801	67	1	different	different	ADJ
ajst-16801	67	2	single	single	ADJ
ajst-16801	67	3	-	-	PUNCT
ajst-16801	67	4	layer	layer	NOUN
ajst-16801	67	5	dictionary	dictionary	ADJ
ajst-16801	67	6	learning	learning	NOUN
ajst-16801	67	7	models	model	NOUN
ajst-16801	67	8	can	can	AUX
ajst-16801	67	9	be	be	AUX
ajst-16801	67	10	constructed	construct	VERB
ajst-16801	67	11	by	by	ADP
ajst-16801	67	12	changing	change	VERB
ajst-16801	67	13	the	the	DET
ajst-16801	67	14	structure	structure	NOUN
ajst-16801	67	15	of	of	ADP
ajst-16801	67	16	𝜓(d	𝜓(d	PROPN
ajst-16801	67	17	,	,	PUNCT
ajst-16801	67	18	x	x	PROPN
ajst-16801	67	19	,	,	PUNCT
ajst-16801	67	20	y	y	PROPN
ajst-16801	67	21	)	)	PUNCT
ajst-16801	67	22	,	,	PUNCT
ajst-16801	67	23	so	so	SCONJ
ajst-16801	67	24	that	that	SCONJ
ajst-16801	67	25	discriminative	discriminative	NOUN
ajst-16801	67	26	dictionaries	dictionary	NOUN
ajst-16801	67	27	and	and	CCONJ
ajst-16801	67	28	sparse	sparse	ADJ
ajst-16801	67	29	representation	representation	NOUN
ajst-16801	67	30	coefficients	coefficient	NOUN
ajst-16801	67	31	adapted	adapt	VERB
ajst-16801	67	32	to	to	ADP
ajst-16801	67	33	different	different	ADJ
ajst-16801	67	34	visual	visual	ADJ
ajst-16801	67	35	tasks	task	NOUN
ajst-16801	67	36	can	can	AUX
ajst-16801	67	37	be	be	AUX
ajst-16801	67	38	learned	learn	VERB
ajst-16801	67	39	.	.	PUNCT
ajst-16801	68	1	2.2	2.2	NUM
ajst-16801	68	2	.	.	PUNCT
ajst-16801	69	1	stack	stack	VERB
ajst-16801	69	2	autoencoder	autoencoder	NOUN
ajst-16801	69	3	the	the	DET
ajst-16801	69	4	stack	stack	NOUN
ajst-16801	69	5	autoencoder	autoencoder	NOUN
ajst-16801	69	6	,	,	PUNCT
ajst-16801	69	7	or	or	CCONJ
ajst-16801	69	8	sae	sae	PROPN
ajst-16801	69	9	,	,	PUNCT
ajst-16801	69	10	is	be	AUX
ajst-16801	69	11	formed	form	VERB
ajst-16801	69	12	by	by	ADP
ajst-16801	69	13	superimposing	superimpose	VERB
ajst-16801	69	14	one	one	NUM
ajst-16801	69	15	ae	ae	PROPN
ajst-16801	69	16	on	on	ADP
ajst-16801	69	17	another	another	DET
ajst-16801	69	18	ae	ae	PROPN
ajst-16801	69	19	,	,	PUNCT
ajst-16801	69	20	and	and	CCONJ
ajst-16801	69	21	an	an	DET
ajst-16801	69	22	autoencoder	autoencoder	NOUN
ajst-16801	69	23	ae	ae	PROPN
ajst-16801	69	24	consists	consist	VERB
ajst-16801	69	25	of	of	ADP
ajst-16801	69	26	two	two	NUM
ajst-16801	69	27	parts	part	NOUN
ajst-16801	69	28	:	:	PUNCT
ajst-16801	69	29	encoding	encode	VERB
ajst-16801	69	30	and	and	CCONJ
ajst-16801	69	31	decoding	decode	VERB
ajst-16801	69	32	.	.	PUNCT
ajst-16801	70	1	the	the	DET
ajst-16801	70	2	autoencoder	autoencoder	NOUN
ajst-16801	70	3	ae	ae	PROPN
ajst-16801	70	4	maps	map	VERB
ajst-16801	70	5	the	the	DET
ajst-16801	70	6	input	input	NOUN
ajst-16801	70	7	to	to	ADP
ajst-16801	70	8	the	the	DET
ajst-16801	70	9	hidden	hide	VERB
ajst-16801	70	10	layer	layer	NOUN
ajst-16801	70	11	,	,	PUNCT
ajst-16801	70	12	while	while	SCONJ
ajst-16801	70	13	the	the	DET
ajst-16801	70	14	decoder	decoder	NOUN
ajst-16801	70	15	maps	map	VERB
ajst-16801	70	16	the	the	DET
ajst-16801	70	17	hidden	hide	VERB
ajst-16801	70	18	layer	layer	NOUN
ajst-16801	70	19	feature	feature	NOUN
ajst-16801	70	20	representation	representation	NOUN
ajst-16801	70	21	back	back	ADV
ajst-16801	70	22	to	to	ADP
ajst-16801	70	23	the	the	DET
ajst-16801	70	24	original	original	ADJ
ajst-16801	70	25	data	datum	NOUN
ajst-16801	70	26	.	.	PUNCT
ajst-16801	71	1	for	for	ADP
ajst-16801	71	2	a	a	DET
ajst-16801	71	3	given	give	VERB
ajst-16801	71	4	input	input	NOUN
ajst-16801	71	5	vector	vector	NOUN
ajst-16801	71	6	𝑥	𝑥	NOUN
ajst-16801	71	7	,	,	PUNCT
ajst-16801	71	8	it	it	PRON
ajst-16801	71	9	tries	try	VERB
ajst-16801	71	10	to	to	PART
ajst-16801	71	11	learn	learn	VERB
ajst-16801	71	12	a	a	DET
ajst-16801	71	13	hidden	hide	VERB
ajst-16801	71	14	layer	layer	NOUN
ajst-16801	71	15	feature	feature	NOUN
ajst-16801	71	16	representing	represent	VERB
ajst-16801	71	17	ℎ	ℎ	NOUN
ajst-16801	71	18	,	,	PUNCT
ajst-16801	71	19	such	such	ADJ
ajst-16801	71	20	that	that	SCONJ
ajst-16801	71	21	ℎ	ℎ	PROPN
ajst-16801	71	22	,	,	PUNCT
ajst-16801	71	23	𝑥	𝑥	PRON
ajst-16801	71	24	forces	force	VERB
ajst-16801	71	25	the	the	DET
ajst-16801	71	26	output	output	NOUN
ajst-16801	71	27	to	to	PART
ajst-16801	71	28	equal	equal	VERB
ajst-16801	71	29	the	the	DET
ajst-16801	71	30	input	input	NOUN
ajst-16801	71	31	,	,	PUNCT
ajst-16801	71	32	where	where	SCONJ
ajst-16801	71	33	w	w	NOUN
ajst-16801	71	34	is	be	AUX
ajst-16801	71	35	the	the	DET
ajst-16801	71	36	network	network	PROPN
ajst-16801	71	37	dictionary	dictionary	PROPN
ajst-16801	71	38	initialization	initialization	PROPN
ajst-16801	71	39	sparse	sparse	ADJ
ajst-16801	71	40	coding	code	VERB
ajst-16801	71	41	solution	solution	NOUN
ajst-16801	71	42	dictionary	dictionary	PROPN
ajst-16801	71	43	update	update	NOUN
ajst-16801	71	44	237	237	NUM
ajst-16801	71	45	weight	weight	NOUN
ajst-16801	71	46	parameter	parameter	NOUN
ajst-16801	71	47	and	and	CCONJ
ajst-16801	71	48	b	b	NOUN
ajst-16801	71	49	is	be	AUX
ajst-16801	71	50	the	the	DET
ajst-16801	71	51	bias	bias	NOUN
ajst-16801	71	52	parameter	parameter	NOUN
ajst-16801	71	53	.	.	PUNCT
ajst-16801	72	1	there	there	PRON
ajst-16801	72	2	are	be	VERB
ajst-16801	72	3	several	several	ADJ
ajst-16801	72	4	extensions	extension	NOUN
ajst-16801	72	5	to	to	ADP
ajst-16801	72	6	the	the	DET
ajst-16801	72	7	basic	basic	ADJ
ajst-16801	72	8	autoencoder	autoencoder	NOUN
ajst-16801	72	9	ae	ae	PROPN
ajst-16801	72	10	architecture	architecture	NOUN
ajst-16801	72	11	,	,	PUNCT
ajst-16801	72	12	one	one	NUM
ajst-16801	72	13	of	of	ADP
ajst-16801	72	14	which	which	PRON
ajst-16801	72	15	is	be	AUX
ajst-16801	72	16	the	the	DET
ajst-16801	72	17	stack	stack	NOUN
ajst-16801	72	18	autoencoder	autoencoder	NOUN
ajst-16801	72	19	sae	sae	PROPN
ajst-16801	72	20	.	.	PUNCT
ajst-16801	73	1	it	it	PRON
ajst-16801	73	2	has	have	VERB
ajst-16801	73	3	multiple	multiple	ADJ
ajst-16801	73	4	hidden	hide	VERB
ajst-16801	73	5	layers	layer	NOUN
ajst-16801	73	6	,	,	PUNCT
ajst-16801	73	7	and	and	CCONJ
ajst-16801	73	8	the	the	DET
ajst-16801	73	9	hidden	hide	VERB
ajst-16801	73	10	layer	layer	NOUN
ajst-16801	73	11	features	feature	NOUN
ajst-16801	73	12	of	of	ADP
ajst-16801	73	13	the	the	DET
ajst-16801	73	14	upper	upper	ADJ
ajst-16801	73	15	layer	layer	NOUN
ajst-16801	73	16	are	be	AUX
ajst-16801	73	17	represented	represent	VERB
ajst-16801	73	18	as	as	ADP
ajst-16801	73	19	the	the	DET
ajst-16801	73	20	input	input	NOUN
ajst-16801	73	21	of	of	ADP
ajst-16801	73	22	the	the	DET
ajst-16801	73	23	current	current	ADJ
ajst-16801	73	24	layer	layer	NOUN
ajst-16801	73	25	,	,	PUNCT
ajst-16801	73	26	which	which	PRON
ajst-16801	73	27	is	be	AUX
ajst-16801	73	28	stacked	stack	VERB
ajst-16801	73	29	layer	layer	NOUN
ajst-16801	73	30	by	by	ADP
ajst-16801	73	31	layer	layer	NOUN
ajst-16801	73	32	to	to	PART
ajst-16801	73	33	form	form	VERB
ajst-16801	73	34	a	a	DET
ajst-16801	73	35	deep	deep	ADJ
ajst-16801	73	36	network	network	NOUN
ajst-16801	73	37	,	,	PUNCT
ajst-16801	73	38	as	as	SCONJ
ajst-16801	73	39	shown	show	VERB
ajst-16801	73	40	in	in	ADP
ajst-16801	73	41	figure	figure	NOUN
ajst-16801	73	42	2	2	NUM
ajst-16801	73	43	.	.	PUNCT
ajst-16801	73	44	figure	figure	NOUN
ajst-16801	73	45	2	2	NUM
ajst-16801	73	46	.	.	PUNCT
ajst-16801	74	1	the	the	DET
ajst-16801	74	2	structure	structure	NOUN
ajst-16801	74	3	of	of	ADP
ajst-16801	74	4	stack	stack	NOUN
ajst-16801	74	5	autoencoder	autoencoder	PROPN
ajst-16801	74	6	sae	sae	PROPN
ajst-16801	74	7	is	be	AUX
ajst-16801	74	8	an	an	DET
ajst-16801	74	9	unsupervised	unsupervised	ADJ
ajst-16801	74	10	training	training	NOUN
ajst-16801	74	11	architecture	architecture	NOUN
ajst-16801	74	12	,	,	PUNCT
ajst-16801	74	13	which	which	PRON
ajst-16801	74	14	has	have	VERB
ajst-16801	74	15	two	two	NUM
ajst-16801	74	16	stages	stage	NOUN
ajst-16801	74	17	,	,	PUNCT
ajst-16801	74	18	encoding	encode	VERB
ajst-16801	74	19	and	and	CCONJ
ajst-16801	74	20	decoding	decode	VERB
ajst-16801	74	21	.	.	PUNCT
ajst-16801	75	1	in	in	ADP
ajst-16801	75	2	the	the	DET
ajst-16801	75	3	coding	code	VERB
ajst-16801	75	4	stage	stage	NOUN
ajst-16801	75	5	,	,	PUNCT
ajst-16801	75	6	each	each	DET
ajst-16801	75	7	layer	layer	NOUN
ajst-16801	75	8	of	of	ADP
ajst-16801	75	9	autoencoder	autoencoder	NOUN
ajst-16801	75	10	needs	need	VERB
ajst-16801	75	11	to	to	PART
ajst-16801	75	12	be	be	AUX
ajst-16801	75	13	encoded	encode	VERB
ajst-16801	75	14	from	from	ADP
ajst-16801	75	15	front	front	NOUN
ajst-16801	75	16	to	to	ADP
ajst-16801	75	17	back	back	ADV
ajst-16801	75	18	along	along	ADP
ajst-16801	75	19	the	the	DET
ajst-16801	75	20	network	network	NOUN
ajst-16801	75	21	,	,	PUNCT
ajst-16801	75	22	while	while	SCONJ
ajst-16801	75	23	in	in	ADP
ajst-16801	75	24	the	the	DET
ajst-16801	75	25	decoding	decode	VERB
ajst-16801	75	26	stage	stage	NOUN
ajst-16801	75	27	,	,	PUNCT
ajst-16801	75	28	each	each	DET
ajst-16801	75	29	layer	layer	NOUN
ajst-16801	75	30	of	of	ADP
ajst-16801	75	31	decoder	decoder	NOUN
ajst-16801	75	32	needs	need	VERB
ajst-16801	75	33	to	to	PART
ajst-16801	75	34	be	be	AUX
ajst-16801	75	35	decoded	decode	VERB
ajst-16801	75	36	from	from	ADP
ajst-16801	75	37	back	back	ADV
ajst-16801	75	38	to	to	ADP
ajst-16801	75	39	front	front	NOUN
ajst-16801	75	40	by	by	ADP
ajst-16801	75	41	the	the	DET
ajst-16801	75	42	network	network	NOUN
ajst-16801	75	43	.	.	PUNCT
ajst-16801	76	1	in	in	ADP
ajst-16801	76	2	the	the	DET
ajst-16801	76	3	stack	stack	NOUN
ajst-16801	76	4	self	self	NOUN
ajst-16801	76	5	-	-	PUNCT
ajst-16801	76	6	coding	code	VERB
ajst-16801	76	7	neural	neural	ADJ
ajst-16801	76	8	network	network	NOUN
ajst-16801	76	9	,	,	PUNCT
ajst-16801	76	10	the	the	DET
ajst-16801	76	11	greedy	greedy	ADJ
ajst-16801	76	12	training	training	NOUN
ajst-16801	76	13	method	method	NOUN
ajst-16801	76	14	is	be	AUX
ajst-16801	76	15	used	use	VERB
ajst-16801	76	16	to	to	PART
ajst-16801	76	17	obtain	obtain	VERB
ajst-16801	76	18	the	the	DET
ajst-16801	76	19	network	network	NOUN
ajst-16801	76	20	parameters	parameter	NOUN
ajst-16801	76	21	,	,	PUNCT
ajst-16801	76	22	that	that	ADV
ajst-16801	76	23	is	is	ADV
ajst-16801	76	24	,	,	PUNCT
ajst-16801	76	25	the	the	DET
ajst-16801	76	26	original	original	ADJ
ajst-16801	76	27	sample	sample	NOUN
ajst-16801	76	28	is	be	AUX
ajst-16801	76	29	first	first	ADJ
ajst-16801	76	30	input	input	NOUN
ajst-16801	76	31	to	to	PART
ajst-16801	76	32	train	train	VERB
ajst-16801	76	33	the	the	DET
ajst-16801	76	34	first	first	ADJ
ajst-16801	76	35	layer	layer	NOUN
ajst-16801	76	36	of	of	ADP
ajst-16801	76	37	the	the	DET
ajst-16801	76	38	neural	neural	ADJ
ajst-16801	76	39	network	network	NOUN
ajst-16801	76	40	,	,	PUNCT
ajst-16801	76	41	so	so	SCONJ
ajst-16801	76	42	as	as	SCONJ
ajst-16801	76	43	to	to	PART
ajst-16801	76	44	obtain	obtain	VERB
ajst-16801	76	45	the	the	DET
ajst-16801	76	46	parameters	parameter	NOUN
ajst-16801	76	47	of	of	ADP
ajst-16801	76	48	the	the	DET
ajst-16801	76	49	first	first	ADJ
ajst-16801	76	50	layer	layer	NOUN
ajst-16801	76	51	network	network	NOUN
ajst-16801	76	52	.	.	PUNCT
ajst-16801	77	1	then	then	ADV
ajst-16801	77	2	the	the	DET
ajst-16801	77	3	original	original	ADJ
ajst-16801	77	4	sample	sample	NOUN
ajst-16801	77	5	is	be	AUX
ajst-16801	77	6	converted	convert	VERB
ajst-16801	77	7	into	into	ADP
ajst-16801	77	8	a	a	DET
ajst-16801	77	9	vector	vector	NOUN
ajst-16801	77	10	x	x	PUNCT
ajst-16801	77	11	composed	compose	VERB
ajst-16801	77	12	of	of	ADP
ajst-16801	77	13	the	the	DET
ajst-16801	77	14	activation	activation	NOUN
ajst-16801	77	15	values	value	NOUN
ajst-16801	77	16	of	of	ADP
ajst-16801	77	17	the	the	DET
ajst-16801	77	18	hidden	hide	VERB
ajst-16801	77	19	layer	layer	NOUN
ajst-16801	77	20	through	through	ADP
ajst-16801	77	21	the	the	DET
ajst-16801	77	22	first	first	ADJ
ajst-16801	77	23	layer	layer	NOUN
ajst-16801	77	24	network	network	NOUN
ajst-16801	77	25	with	with	ADP
ajst-16801	77	26	the	the	DET
ajst-16801	77	27	parameters	parameter	NOUN
ajst-16801	77	28	determined	determine	VERB
ajst-16801	77	29	,	,	PUNCT
ajst-16801	77	30	and	and	CCONJ
ajst-16801	77	31	x	x	X
ajst-16801	77	32	is	be	AUX
ajst-16801	77	33	continued	continue	VERB
ajst-16801	77	34	as	as	ADP
ajst-16801	77	35	the	the	DET
ajst-16801	77	36	input	input	NOUN
ajst-16801	77	37	of	of	ADP
ajst-16801	77	38	the	the	DET
ajst-16801	77	39	second	second	ADJ
ajst-16801	77	40	layer	layer	NOUN
ajst-16801	77	41	of	of	ADP
ajst-16801	77	42	the	the	DET
ajst-16801	77	43	neural	neural	ADJ
ajst-16801	77	44	network	network	NOUN
ajst-16801	77	45	.	.	PUNCT
ajst-16801	78	1	the	the	DET
ajst-16801	78	2	parameters	parameter	NOUN
ajst-16801	78	3	of	of	ADP
ajst-16801	78	4	the	the	DET
ajst-16801	78	5	second	second	ADJ
ajst-16801	78	6	layer	layer	NOUN
ajst-16801	78	7	network	network	NOUN
ajst-16801	78	8	are	be	AUX
ajst-16801	78	9	obtained	obtain	VERB
ajst-16801	78	10	by	by	ADP
ajst-16801	78	11	training	training	NOUN
ajst-16801	78	12	.	.	PUNCT
ajst-16801	79	1	similarly	similarly	ADV
ajst-16801	79	2	,	,	PUNCT
ajst-16801	79	3	the	the	DET
ajst-16801	79	4	parameters	parameter	NOUN
ajst-16801	79	5	of	of	ADP
ajst-16801	79	6	each	each	DET
ajst-16801	79	7	subsequent	subsequent	ADJ
ajst-16801	79	8	layer	layer	NOUN
ajst-16801	79	9	of	of	ADP
ajst-16801	79	10	the	the	DET
ajst-16801	79	11	network	network	NOUN
ajst-16801	79	12	are	be	AUX
ajst-16801	79	13	trained	train	VERB
ajst-16801	79	14	in	in	ADP
ajst-16801	79	15	this	this	DET
ajst-16801	79	16	way	way	NOUN
ajst-16801	79	17	.	.	PUNCT
ajst-16801	80	1	generally	generally	ADV
ajst-16801	80	2	speaking	speak	VERB
ajst-16801	80	3	,	,	PUNCT
ajst-16801	80	4	in	in	ADP
ajst-16801	80	5	order	order	NOUN
ajst-16801	80	6	to	to	PART
ajst-16801	80	7	obtain	obtain	VERB
ajst-16801	80	8	better	well	ADJ
ajst-16801	80	9	training	training	NOUN
ajst-16801	80	10	results	result	NOUN
ajst-16801	80	11	,	,	PUNCT
ajst-16801	80	12	fine	fine	ADV
ajst-16801	80	13	-	-	PUNCT
ajst-16801	80	14	tuning	tune	VERB
ajst-16801	80	15	steps	step	NOUN
ajst-16801	80	16	need	need	VERB
ajst-16801	80	17	to	to	PART
ajst-16801	80	18	be	be	AUX
ajst-16801	80	19	added	add	VERB
ajst-16801	80	20	to	to	ADP
ajst-16801	80	21	the	the	DET
ajst-16801	80	22	above	above	ADJ
ajst-16801	80	23	training	training	NOUN
ajst-16801	80	24	process	process	NOUN
ajst-16801	80	25	,	,	PUNCT
ajst-16801	80	26	that	that	ADV
ajst-16801	80	27	is	is	ADV
ajst-16801	80	28	,	,	PUNCT
ajst-16801	80	29	after	after	SCONJ
ajst-16801	80	30	the	the	DET
ajst-16801	80	31	training	training	NOUN
ajst-16801	80	32	process	process	NOUN
ajst-16801	80	33	is	be	AUX
ajst-16801	80	34	completed	complete	VERB
ajst-16801	80	35	,	,	PUNCT
ajst-16801	80	36	the	the	DET
ajst-16801	80	37	backpropagation	backpropagation	NOUN
ajst-16801	80	38	method	method	NOUN
ajst-16801	80	39	is	be	AUX
ajst-16801	80	40	used	use	VERB
ajst-16801	80	41	to	to	PART
ajst-16801	80	42	optimize	optimize	VERB
ajst-16801	80	43	and	and	CCONJ
ajst-16801	80	44	adjust	adjust	VERB
ajst-16801	80	45	the	the	DET
ajst-16801	80	46	parameters	parameter	NOUN
ajst-16801	80	47	of	of	ADP
ajst-16801	80	48	each	each	DET
ajst-16801	80	49	layer	layer	NOUN
ajst-16801	80	50	,	,	PUNCT
ajst-16801	80	51	because	because	SCONJ
ajst-16801	80	52	the	the	DET
ajst-16801	80	53	training	training	NOUN
ajst-16801	80	54	of	of	ADP
ajst-16801	80	55	each	each	DET
ajst-16801	80	56	layer	layer	NOUN
ajst-16801	80	57	is	be	AUX
ajst-16801	80	58	to	to	PART
ajst-16801	80	59	obtain	obtain	VERB
ajst-16801	80	60	the	the	DET
ajst-16801	80	61	parameters	parameter	NOUN
ajst-16801	80	62	of	of	ADP
ajst-16801	80	63	the	the	DET
ajst-16801	80	64	current	current	ADJ
ajst-16801	80	65	layer	layer	NOUN
ajst-16801	80	66	by	by	ADP
ajst-16801	80	67	fixing	fix	VERB
ajst-16801	80	68	the	the	DET
ajst-16801	80	69	parameters	parameter	NOUN
ajst-16801	80	70	of	of	ADP
ajst-16801	80	71	other	other	ADJ
ajst-16801	80	72	layers	layer	NOUN
ajst-16801	80	73	,	,	PUNCT
ajst-16801	80	74	and	and	CCONJ
ajst-16801	80	75	the	the	DET
ajst-16801	80	76	parameters	parameter	NOUN
ajst-16801	80	77	at	at	ADP
ajst-16801	80	78	this	this	DET
ajst-16801	80	79	time	time	NOUN
ajst-16801	80	80	are	be	AUX
ajst-16801	80	81	not	not	PART
ajst-16801	80	82	optimal	optimal	ADJ
ajst-16801	80	83	for	for	ADP
ajst-16801	80	84	the	the	DET
ajst-16801	80	85	whole	whole	ADJ
ajst-16801	80	86	network	network	NOUN
ajst-16801	80	87	.	.	PUNCT
ajst-16801	81	1	in	in	ADP
ajst-16801	81	2	practice	practice	NOUN
ajst-16801	81	3	,	,	PUNCT
ajst-16801	81	4	the	the	DET
ajst-16801	81	5	fine	fine	ADV
ajst-16801	81	6	-	-	PUNCT
ajst-16801	81	7	tuning	tuning	NOUN
ajst-16801	81	8	method	method	NOUN
ajst-16801	81	9	is	be	AUX
ajst-16801	81	10	generally	generally	ADV
ajst-16801	81	11	used	use	VERB
ajst-16801	81	12	when	when	SCONJ
ajst-16801	81	13	the	the	DET
ajst-16801	81	14	parameters	parameter	NOUN
ajst-16801	81	15	are	be	AUX
ajst-16801	81	16	trained	train	VERB
ajst-16801	81	17	close	close	ADV
ajst-16801	81	18	to	to	ADP
ajst-16801	81	19	the	the	DET
ajst-16801	81	20	convergence	convergence	NOUN
ajst-16801	81	21	state	state	NOUN
ajst-16801	81	22	.	.	PUNCT
ajst-16801	82	1	if	if	SCONJ
ajst-16801	82	2	the	the	DET
ajst-16801	82	3	randomly	randomly	ADV
ajst-16801	82	4	initialized	initialize	VERB
ajst-16801	82	5	weight	weight	NOUN
ajst-16801	82	6	parameters	parameter	NOUN
ajst-16801	82	7	are	be	AUX
ajst-16801	82	8	fine	fine	ADV
ajst-16801	82	9	-	-	PUNCT
ajst-16801	82	10	tuned	tune	VERB
ajst-16801	82	11	directly	directly	ADV
ajst-16801	82	12	,	,	PUNCT
ajst-16801	82	13	the	the	DET
ajst-16801	82	14	results	result	NOUN
ajst-16801	82	15	will	will	AUX
ajst-16801	82	16	be	be	AUX
ajst-16801	82	17	poor	poor	ADJ
ajst-16801	82	18	because	because	SCONJ
ajst-16801	82	19	the	the	DET
ajst-16801	82	20	parameters	parameter	NOUN
ajst-16801	82	21	only	only	ADV
ajst-16801	82	22	converge	converge	VERB
ajst-16801	82	23	to	to	ADP
ajst-16801	82	24	the	the	DET
ajst-16801	82	25	local	local	ADJ
ajst-16801	82	26	optimal	optimal	ADJ
ajst-16801	82	27	.	.	PUNCT
ajst-16801	83	1	ddl	ddl	PROPN
ajst-16801	83	2	is	be	AUX
ajst-16801	83	3	similar	similar	ADJ
ajst-16801	83	4	to	to	AUX
ajst-16801	83	5	sae	sae	VERB
ajst-16801	83	6	in	in	ADP
ajst-16801	83	7	coding	code	VERB
ajst-16801	83	8	part	part	NOUN
ajst-16801	83	9	but	but	CCONJ
ajst-16801	83	10	in	in	ADP
ajst-16801	83	11	opposite	opposite	ADJ
ajst-16801	83	12	training	training	NOUN
ajst-16801	83	13	direction	direction	NOUN
ajst-16801	83	14	.	.	PUNCT
ajst-16801	84	1	ddl	ddl	PROPN
ajst-16801	84	2	reconstructs	reconstruct	VERB
ajst-16801	84	3	the	the	DET
ajst-16801	84	4	original	original	ADJ
ajst-16801	84	5	signal	signal	NOUN
ajst-16801	84	6	x	x	SYM
ajst-16801	84	7	=	=	NOUN
ajst-16801	84	8	d*z	d*z	NOUN
ajst-16801	84	9	by	by	ADP
ajst-16801	84	10	learning	learn	VERB
ajst-16801	84	11	dictionary	dictionary	ADJ
ajst-16801	84	12	d	d	NOUN
ajst-16801	84	13	and	and	CCONJ
ajst-16801	84	14	depth	depth	NOUN
ajst-16801	84	15	feature	feature	NOUN
ajst-16801	84	16	z.	z.	PROPN
ajst-16801	84	17	sae	sae	PROPN
ajst-16801	84	18	learns	learn	VERB
ajst-16801	84	19	the	the	DET
ajst-16801	84	20	weight	weight	NOUN
ajst-16801	84	21	parameter	parameter	NOUN
ajst-16801	84	22	w	w	NOUN
ajst-16801	85	1	so	so	SCONJ
ajst-16801	85	2	that	that	SCONJ
ajst-16801	85	3	w*x	w*x	PROPN
ajst-16801	85	4	=	=	NOUN
ajst-16801	85	5	z	z	NOUN
ajst-16801	85	6	to	to	PART
ajst-16801	85	7	obtain	obtain	VERB
ajst-16801	85	8	the	the	DET
ajst-16801	85	9	output	output	NOUN
ajst-16801	85	10	signal	signal	NOUN
ajst-16801	85	11	,	,	PUNCT
ajst-16801	85	12	where	where	SCONJ
ajst-16801	85	13	x	x	PRON
ajst-16801	85	14	is	be	AUX
ajst-16801	85	15	the	the	DET
ajst-16801	85	16	feature	feature	NOUN
ajst-16801	85	17	representation	representation	NOUN
ajst-16801	85	18	and	and	CCONJ
ajst-16801	85	19	z	z	NOUN
ajst-16801	85	20	is	be	AUX
ajst-16801	85	21	the	the	DET
ajst-16801	85	22	output	output	NOUN
ajst-16801	85	23	signal	signal	NOUN
ajst-16801	85	24	.	.	PUNCT
ajst-16801	86	1	2.3	2.3	NUM
ajst-16801	86	2	.	.	X
ajst-16801	87	1	2.3	2.3	NUM
ajst-16801	87	2	deep	deep	ADJ
ajst-16801	87	3	convolutional	convolutional	ADJ
ajst-16801	87	4	dictionary	dictionary	ADJ
ajst-16801	87	5	learning	learning	NOUN
ajst-16801	87	6	model	model	NOUN
ajst-16801	87	7	by	by	ADP
ajst-16801	87	8	combining	combine	VERB
ajst-16801	87	9	the	the	DET
ajst-16801	87	10	dictionary	dictionary	ADJ
ajst-16801	87	11	learning	learning	NOUN
ajst-16801	87	12	model	model	NOUN
ajst-16801	87	13	with	with	ADP
ajst-16801	87	14	the	the	DET
ajst-16801	87	15	deep	deep	ADJ
ajst-16801	87	16	neural	neural	ADJ
ajst-16801	87	17	network	network	NOUN
ajst-16801	87	18	,	,	PUNCT
ajst-16801	87	19	the	the	DET
ajst-16801	87	20	deep	deep	ADJ
ajst-16801	87	21	potential	potential	ADJ
ajst-16801	87	22	representation	representation	NOUN
ajst-16801	87	23	of	of	ADP
ajst-16801	87	24	the	the	DET
ajst-16801	87	25	original	original	ADJ
ajst-16801	87	26	image	image	NOUN
ajst-16801	87	27	data	datum	NOUN
ajst-16801	87	28	becomes	become	VERB
ajst-16801	87	29	the	the	DET
ajst-16801	87	30	input	input	NOUN
ajst-16801	87	31	of	of	ADP
ajst-16801	87	32	the	the	DET
ajst-16801	87	33	dictionary	dictionary	ADJ
ajst-16801	87	34	learning	learning	PROPN
ajst-16801	87	35	model	model	NOUN
ajst-16801	87	36	,	,	PUNCT
ajst-16801	87	37	which	which	PRON
ajst-16801	87	38	solves	solve	VERB
ajst-16801	87	39	the	the	DET
ajst-16801	87	40	spatial	spatial	ADJ
ajst-16801	87	41	redundancy	redundancy	NOUN
ajst-16801	87	42	problem	problem	NOUN
ajst-16801	87	43	of	of	ADP
ajst-16801	87	44	the	the	DET
ajst-16801	87	45	feature	feature	NOUN
ajst-16801	87	46	map	map	NOUN
ajst-16801	87	47	,	,	PUNCT
ajst-16801	87	48	and	and	CCONJ
ajst-16801	87	49	helps	help	VERB
ajst-16801	87	50	to	to	PART
ajst-16801	87	51	improve	improve	VERB
ajst-16801	87	52	the	the	DET
ajst-16801	87	53	sparse	sparse	ADJ
ajst-16801	87	54	reconstruction	reconstruction	NOUN
ajst-16801	87	55	of	of	ADP
ajst-16801	87	56	data	datum	NOUN
ajst-16801	87	57	and	and	CCONJ
ajst-16801	87	58	more	more	ADV
ajst-16801	87	59	robust	robust	ADJ
ajst-16801	87	60	feature	feature	NOUN
ajst-16801	87	61	description	description	NOUN
ajst-16801	87	62	under	under	ADP
ajst-16801	87	63	small	small	ADJ
ajst-16801	87	64	samples	sample	NOUN
ajst-16801	87	65	.	.	PUNCT
ajst-16801	88	1	therefore	therefore	ADV
ajst-16801	88	2	,	,	PUNCT
ajst-16801	88	3	a	a	DET
ajst-16801	88	4	new	new	ADJ
ajst-16801	88	5	dictionary	dictionary	ADJ
ajst-16801	88	6	learning	learning	NOUN
ajst-16801	88	7	method	method	NOUN
ajst-16801	88	8	called	call	VERB
ajst-16801	88	9	deep	deep	ADV
ajst-16801	88	10	sparse	sparse	ADJ
ajst-16801	88	11	dictionary	dictionary	ADJ
ajst-16801	88	12	learning	learning	NOUN
ajst-16801	88	13	(	(	PUNCT
ajst-16801	88	14	dsdl	dsdl	PROPN
ajst-16801	88	15	)	)	PUNCT
ajst-16801	88	16	is	be	AUX
ajst-16801	88	17	proposed	propose	VERB
ajst-16801	88	18	.	.	PUNCT
ajst-16801	89	1	stack	stack	NOUN
ajst-16801	89	2	autoencoders	autoencoder	NOUN
ajst-16801	89	3	are	be	AUX
ajst-16801	89	4	widely	widely	ADV
ajst-16801	89	5	used	use	VERB
ajst-16801	89	6	to	to	PART
ajst-16801	89	7	extract	extract	VERB
ajst-16801	89	8	depth	depth	NOUN
ajst-16801	89	9	features	feature	NOUN
ajst-16801	89	10	from	from	ADP
ajst-16801	89	11	data	datum	NOUN
ajst-16801	89	12	in	in	ADP
ajst-16801	89	13	an	an	DET
ajst-16801	89	14	unsupervised	unsupervised	ADJ
ajst-16801	89	15	manner	manner	NOUN
ajst-16801	89	16	.	.	PUNCT
ajst-16801	90	1	by	by	ADP
ajst-16801	90	2	designing	design	VERB
ajst-16801	90	3	a	a	DET
ajst-16801	90	4	certain	certain	ADJ
ajst-16801	90	5	encoder	encoder	NOUN
ajst-16801	90	6	layer	layer	NOUN
ajst-16801	90	7	and	and	CCONJ
ajst-16801	90	8	a	a	DET
ajst-16801	90	9	decoder	decoder	NOUN
ajst-16801	90	10	layer	layer	NOUN
ajst-16801	90	11	,	,	PUNCT
ajst-16801	90	12	the	the	DET
ajst-16801	90	13	reconstruction	reconstruction	NOUN
ajst-16801	90	14	error	error	NOUN
ajst-16801	90	15	between	between	ADP
ajst-16801	90	16	the	the	DET
ajst-16801	90	17	decoded	decode	VERB
ajst-16801	90	18	output	output	NOUN
ajst-16801	90	19	and	and	CCONJ
ajst-16801	90	20	the	the	DET
ajst-16801	90	21	original	original	ADJ
ajst-16801	90	22	input	input	NOUN
ajst-16801	90	23	is	be	AUX
ajst-16801	90	24	minimized	minimize	VERB
ajst-16801	90	25	,	,	PUNCT
ajst-16801	90	26	so	so	CCONJ
ajst-16801	90	27	it	it	PRON
ajst-16801	90	28	is	be	AUX
ajst-16801	90	29	often	often	ADV
ajst-16801	90	30	used	use	VERB
ajst-16801	90	31	for	for	ADP
ajst-16801	90	32	data	datum	NOUN
ajst-16801	90	33	compression	compression	NOUN
ajst-16801	90	34	and	and	CCONJ
ajst-16801	90	35	denoising	denoising	NOUN
ajst-16801	90	36	tasks	task	NOUN
ajst-16801	90	37	.	.	PUNCT
ajst-16801	91	1	the	the	DET
ajst-16801	91	2	deep	deep	ADJ
ajst-16801	91	3	sparse	sparse	ADJ
ajst-16801	91	4	dictionary	dictionary	ADJ
ajst-16801	91	5	learning	learning	NOUN
ajst-16801	91	6	method	method	NOUN
ajst-16801	91	7	proposed	propose	VERB
ajst-16801	91	8	in	in	ADP
ajst-16801	91	9	this	this	DET
ajst-16801	91	10	paper	paper	NOUN
ajst-16801	91	11	uses	use	VERB
ajst-16801	91	12	convolution	convolution	NOUN
ajst-16801	91	13	operators	operator	NOUN
ajst-16801	91	14	to	to	PART
ajst-16801	91	15	simultaneously	simultaneously	ADV
ajst-16801	91	16	learn	learn	VERB
ajst-16801	91	17	depth	depth	NOUN
ajst-16801	91	18	features	feature	NOUN
ajst-16801	91	19	and	and	CCONJ
ajst-16801	91	20	corresponding	correspond	VERB
ajst-16801	91	21	discriminant	discriminant	ADJ
ajst-16801	91	22	dictionary	dictionary	ADJ
ajst-16801	91	23	pairs	pair	NOUN
ajst-16801	91	24	.	.	PUNCT
ajst-16801	92	1	as	as	SCONJ
ajst-16801	92	2	can	can	AUX
ajst-16801	92	3	be	be	AUX
ajst-16801	92	4	seen	see	VERB
ajst-16801	92	5	from	from	ADP
ajst-16801	92	6	figure	figure	NOUN
ajst-16801	92	7	3	3	NUM
ajst-16801	92	8	,	,	PUNCT
ajst-16801	92	9	the	the	DET
ajst-16801	92	10	overall	overall	ADJ
ajst-16801	92	11	architecture	architecture	NOUN
ajst-16801	92	12	is	be	AUX
ajst-16801	92	13	as	as	SCONJ
ajst-16801	92	14	follows	follow	VERB
ajst-16801	92	15	:	:	PUNCT
ajst-16801	92	16	1	1	X
ajst-16801	92	17	.	.	X
ajst-16801	92	18	stack	stack	VERB
ajst-16801	92	19	autoencoder	autoencoder	NOUN
ajst-16801	92	20	.	.	PUNCT
ajst-16801	93	1	the	the	DET
ajst-16801	93	2	input	input	NOUN
ajst-16801	93	3	training	training	NOUN
ajst-16801	93	4	image	image	NOUN
ajst-16801	93	5	y	y	PROPN
ajst-16801	93	6	is	be	AUX
ajst-16801	93	7	sent	send	VERB
ajst-16801	93	8	to	to	ADP
ajst-16801	93	9	the	the	DET
ajst-16801	93	10	encoder	encoder	NOUN
ajst-16801	93	11	layer	layer	NOUN
ajst-16801	93	12	to	to	PART
ajst-16801	93	13	form	form	VERB
ajst-16801	93	14	a	a	DET
ajst-16801	93	15	depth	depth	NOUN
ajst-16801	93	16	feature	feature	NOUN
ajst-16801	93	17	z=	z=	PUNCT
ajst-16801	93	18	h(y;𝜃	h(y;𝜃	PROPN
ajst-16801	93	19	)	)	PUNCT
ajst-16801	93	20	,	,	PUNCT
ajst-16801	93	21	and	and	CCONJ
ajst-16801	93	22	then	then	ADV
ajst-16801	93	23	the	the	DET
ajst-16801	93	24	reconstructed	reconstructed	ADJ
ajst-16801	93	25	𝑌	𝑌	PROPN
ajst-16801	93	26	=	=	PUNCT
ajst-16801	93	27	g(𝑍	g(𝑍	ADV
ajst-16801	93	28	;	;	PUNCT
ajst-16801	93	29	𝜃	𝜃	X
ajst-16801	93	30	)	)	PUNCT
ajst-16801	93	31	,	,	PUNCT
ajst-16801	93	32	where	where	SCONJ
ajst-16801	93	33	𝜃	𝜃	NOUN
ajst-16801	93	34	and	and	CCONJ
ajst-16801	93	35	𝜃	𝜃	X
ajst-16801	93	36	are	be	AUX
ajst-16801	93	37	the	the	DET
ajst-16801	93	38	encoder	encoder	NOUN
ajst-16801	93	39	and	and	CCONJ
ajst-16801	93	40	decoder	decoder	NOUN
ajst-16801	93	41	parameters	parameter	NOUN
ajst-16801	93	42	,	,	PUNCT
ajst-16801	93	43	respectively	respectively	ADV
ajst-16801	93	44	,	,	PUNCT
ajst-16801	93	45	and	and	CCONJ
ajst-16801	93	46	h	h	NOUN
ajst-16801	93	47	and	and	CCONJ
ajst-16801	93	48	g	g	PROPN
ajst-16801	93	49	are	be	AUX
ajst-16801	93	50	the	the	DET
ajst-16801	93	51	mapping	mapping	NOUN
ajst-16801	93	52	functions	function	NOUN
ajst-16801	93	53	implemented	implement	VERB
ajst-16801	93	54	by	by	ADP
ajst-16801	93	55	the	the	DET
ajst-16801	93	56	convolutional	convolutional	ADJ
ajst-16801	93	57	and	and	CCONJ
ajst-16801	93	58	deconvolution	deconvolution	NOUN
ajst-16801	93	59	layers	layer	NOUN
ajst-16801	93	60	,	,	PUNCT
ajst-16801	93	61	respectively	respectively	ADV
ajst-16801	93	62	.	.	PUNCT
ajst-16801	94	1	2	2	X
ajst-16801	94	2	.	.	X
ajst-16801	94	3	discriminating	discriminate	VERB
ajst-16801	94	4	dictionary	dictionary	ADJ
ajst-16801	94	5	pair	pair	NOUN
ajst-16801	94	6	learning	learning	NOUN
ajst-16801	94	7	.	.	PUNCT
ajst-16801	95	1	the	the	DET
ajst-16801	95	2	first	first	PROPN
ajst-16801	95	3	dictionary	dictionary	PROPN
ajst-16801	95	4	,	,	PUNCT
ajst-16801	95	5	p	p	PRON
ajst-16801	95	6	,	,	PUNCT
ajst-16801	95	7	serves	serve	VERB
ajst-16801	95	8	as	as	ADP
ajst-16801	95	9	an	an	DET
ajst-16801	95	10	analysis	analysis	NOUN
ajst-16801	95	11	dictionary	dictionary	ADJ
ajst-16801	95	12	to	to	PART
ajst-16801	95	13	get	get	VERB
ajst-16801	95	14	a	a	DET
ajst-16801	95	15	relatively	relatively	ADV
ajst-16801	95	16	low	low	ADJ
ajst-16801	95	17	-	-	PUNCT
ajst-16801	95	18	dimensional	dimensional	ADJ
ajst-16801	95	19	representation	representation	NOUN
ajst-16801	95	20	,	,	PUNCT
ajst-16801	95	21	and	and	CCONJ
ajst-16801	95	22	the	the	DET
ajst-16801	95	23	second	second	ADJ
ajst-16801	95	24	dictionary	dictionary	NOUN
ajst-16801	95	25	,	,	PUNCT
ajst-16801	95	26	d	d	PROPN
ajst-16801	95	27	,	,	PUNCT
ajst-16801	95	28	trains	train	VERB
ajst-16801	95	29	the	the	DET
ajst-16801	95	30	construction	construction	NOUN
ajst-16801	95	31	of	of	ADP
ajst-16801	95	32	the	the	DET
ajst-16801	95	33	image	image	NOUN
ajst-16801	95	34	.	.	PUNCT
ajst-16801	96	1	it	it	PRON
ajst-16801	96	2	is	be	AUX
ajst-16801	96	3	worth	worth	ADJ
ajst-16801	96	4	noting	note	VERB
ajst-16801	96	5	that	that	SCONJ
ajst-16801	96	6	they	they	PRON
ajst-16801	96	7	are	be	AUX
ajst-16801	96	8	all	all	ADV
ajst-16801	96	9	structured	structure	VERB
ajst-16801	96	10	dictionaries	dictionary	NOUN
ajst-16801	96	11	with	with	ADP
ajst-16801	96	12	columns	column	NOUN
ajst-16801	96	13	that	that	PRON
ajst-16801	96	14	are	be	AUX
ajst-16801	96	15	consistent	consistent	ADJ
ajst-16801	96	16	with	with	ADP
ajst-16801	96	17	labels	label	NOUN
ajst-16801	96	18	.	.	PUNCT
ajst-16801	97	1	figure	figure	NOUN
ajst-16801	97	2	3	3	NUM
ajst-16801	97	3	.	.	PUNCT
ajst-16801	98	1	deep	deep	ADJ
ajst-16801	98	2	sparse	sparse	ADJ
ajst-16801	98	3	dictionary	dictionary	ADJ
ajst-16801	98	4	learning	learning	NOUN
ajst-16801	98	5	model	model	NOUN
ajst-16801	98	6	the	the	DET
ajst-16801	98	7	dsdl	dsdl	NOUN
ajst-16801	98	8	proposed	propose	VERB
ajst-16801	98	9	in	in	ADP
ajst-16801	98	10	this	this	DET
ajst-16801	98	11	paper	paper	NOUN
ajst-16801	98	12	aims	aim	VERB
ajst-16801	98	13	to	to	PART
ajst-16801	98	14	train	train	VERB
ajst-16801	98	15	both	both	DET
ajst-16801	98	16	depth	depth	NOUN
ajst-16801	98	17	features	feature	NOUN
ajst-16801	98	18	and	and	CCONJ
ajst-16801	98	19	sparse	sparse	ADJ
ajst-16801	98	20	dictionaries	dictionary	NOUN
ajst-16801	98	21	.	.	PUNCT
ajst-16801	99	1	therefore	therefore	ADV
ajst-16801	99	2	,	,	PUNCT
ajst-16801	99	3	the	the	DET
ajst-16801	99	4	objective	objective	ADJ
ajst-16801	99	5	function	function	NOUN
ajst-16801	99	6	of	of	ADP
ajst-16801	99	7	dsdl	dsdl	PROPN
ajst-16801	99	8	is	be	AUX
ajst-16801	99	9	shown	show	VERB
ajst-16801	99	10	in	in	ADP
ajst-16801	99	11	equation	equation	NOUN
ajst-16801	99	12	(	(	PUNCT
ajst-16801	99	13	4).where	4).where	NOUN
ajst-16801	99	14	z	z	VERB
ajst-16801	99	15	represents	represent	VERB
ajst-16801	99	16	a	a	DET
ajst-16801	99	17	submatrix	submatrix	NOUN
ajst-16801	99	18	that	that	PRON
ajst-16801	99	19	does	do	AUX
ajst-16801	99	20	not	not	PART
ajst-16801	99	21	include	include	VERB
ajst-16801	99	22	the	the	DET
ajst-16801	99	23	kth	kth	PROPN
ajst-16801	99	24	z.	z.	PROPN
ajst-16801	99	25	the	the	DET
ajst-16801	99	26	first	first	ADJ
ajst-16801	99	27	is	be	AUX
ajst-16801	99	28	the	the	DET
ajst-16801	99	29	autoencoder	autoencoder	NOUN
ajst-16801	99	30	loss	loss	NOUN
ajst-16801	99	31	,	,	PUNCT
ajst-16801	99	32	which	which	PRON
ajst-16801	99	33	minimizes	minimize	VERB
ajst-16801	99	34	the	the	DET
ajst-16801	99	35	error	error	NOUN
ajst-16801	99	36	between	between	ADP
ajst-16801	99	37	the	the	DET
ajst-16801	99	38	original	original	ADJ
ajst-16801	99	39	training	training	NOUN
ajst-16801	99	40	image	image	NOUN
ajst-16801	99	41	and	and	CCONJ
ajst-16801	99	42	the	the	DET
ajst-16801	99	43	reconstructed	reconstructed	ADJ
ajst-16801	99	44	image	image	NOUN
ajst-16801	99	45	.	.	PUNCT
ajst-16801	100	1	sparse	sparse	ADJ
ajst-16801	100	2	reconstruction	reconstruction	NOUN
ajst-16801	100	3	is	be	AUX
ajst-16801	100	4	performed	perform	VERB
ajst-16801	100	5	by	by	ADP
ajst-16801	100	6	passing	pass	VERB
ajst-16801	100	7	the	the	DET
ajst-16801	100	8	depth	depth	NOUN
ajst-16801	100	9	feature	feature	NOUN
ajst-16801	100	10	z	z	NOUN
ajst-16801	100	11	of	of	ADP
ajst-16801	100	12	the	the	DET
ajst-16801	100	13	training	training	NOUN
ajst-16801	100	14	image	image	NOUN
ajst-16801	100	15	to	to	ADP
ajst-16801	100	16	the	the	DET
ajst-16801	100	17	first	first	ADJ
ajst-16801	100	18	analysis	analysis	NOUN
ajst-16801	100	19	dictionary	dictionary	ADJ
ajst-16801	100	20	p	p	NOUN
ajst-16801	100	21	for	for	ADP
ajst-16801	100	22	potential	potential	ADJ
ajst-16801	100	23	representation	representation	NOUN
ajst-16801	100	24	of	of	ADP
ajst-16801	100	25	x	x	X
ajst-16801	100	26	,	,	PUNCT
ajst-16801	100	27	and	and	CCONJ
ajst-16801	100	28	subsequently	subsequently	ADV
ajst-16801	100	29	to	to	ADP
ajst-16801	100	30	the	the	DET
ajst-16801	100	31	discriminant	discriminant	NOUN
ajst-16801	100	32	construction	construction	PROPN
ajst-16801	100	33	dictionary	dictionary	PROPN
ajst-16801	100	34	d.	d.	PROPN
ajst-16801	100	35	min	min	PROPN
ajst-16801	100	36	,	,	PUNCT
ajst-16801	100	37	,	,	PUNCT
ajst-16801	100	38	,	,	PUNCT
ajst-16801	100	39	∑	∑	PUNCT
ajst-16801	100	40	∥	∥	PUNCT
ajst-16801	100	41	𝑌	𝑌	PROPN
ajst-16801	100	42	𝑌	𝑌	PROPN
ajst-16801	100	43	∥	∥	PUNCT
ajst-16801	100	44	𝜆	𝜆	SYM
ajst-16801	100	45	∥	∥	SYM
ajst-16801	100	46	𝐷	𝐷	NOUN
ajst-16801	100	47	𝑋	𝑋	NOUN
ajst-16801	100	48	𝑍	𝑍	VERB
ajst-16801	100	49	∥	∥	X
ajst-16801	100	50	∥	∥	PUNCT
ajst-16801	100	51	𝑃	𝑃	NOUN
ajst-16801	100	52	𝑍	𝑍	NOUN
ajst-16801	100	53	𝑋	𝑋	NOUN
ajst-16801	100	54	∥	∥	PUNCT
ajst-16801	100	55	𝜆	𝜆	SYM
ajst-16801	100	56	∥	∥	PUNCT
ajst-16801	100	57	𝑃	𝑃	NOUN
ajst-16801	100	58	�	�	NOUN
ajst-16801	100	59	̅	̅	NOUN
ajst-16801	100	60	�	�	NOUN
ajst-16801	100	61	∥	∥	X
ajst-16801	100	62	,	,	PUNCT
ajst-16801	100	63	(	(	PUNCT
ajst-16801	100	64	4	4	X
ajst-16801	100	65	)	)	PUNCT
ajst-16801	100	66	s.t	s.t	PROPN
ajst-16801	100	67	.	.	PROPN
ajst-16801	101	1	z	z	X
ajst-16801	101	2	=	=	X
ajst-16801	101	3	h(y;𝜃	h(y;𝜃	PROPN
ajst-16801	101	4	)	)	PUNCT
ajst-16801	101	5	,	,	PUNCT
ajst-16801	101	6	𝑌=g(𝑍	𝑌=g(𝑍	PROPN
ajst-16801	101	7	;	;	PUNCT
ajst-16801	101	8	𝜃	𝜃	X
ajst-16801	101	9	,	,	PUNCT
ajst-16801	101	10	𝜃	𝜃	PROPN
ajst-16801	101	11	𝜃	𝜃	NOUN
ajst-16801	101	12	,	,	PUNCT
ajst-16801	101	13	𝜃	𝜃	INTJ
ajst-16801	101	14	,	,	PUNCT
ajst-16801	101	15	in	in	ADP
ajst-16801	101	16	equation	equation	NOUN
ajst-16801	101	17	(	(	PUNCT
ajst-16801	101	18	4	4	NUM
ajst-16801	101	19	)	)	PUNCT
ajst-16801	101	20	,	,	PUNCT
ajst-16801	101	21	the	the	DET
ajst-16801	101	22	parameters	parameter	NOUN
ajst-16801	101	23	λ_1	λ_1	VERB
ajst-16801	101	24	and	and	CCONJ
ajst-16801	101	25	λ_2	λ_2	PROPN
ajst-16801	101	26	maintain	maintain	VERB
ajst-16801	101	27	a	a	DET
ajst-16801	101	28	…	…	PUNCT
ajst-16801	101	29	𝑃	𝑃	NOUN
ajst-16801	101	30	encoders	encoder	NOUN
ajst-16801	101	31	𝑃	𝑃	VERB
ajst-16801	101	32	𝑃	𝑃	NOUN
ajst-16801	101	33	𝐷	𝐷	NOUN
ajst-16801	101	34	decoders	decoder	NOUN
ajst-16801	101	35	input	input	NOUN
ajst-16801	102	1	o	o	NOUN
ajst-16801	102	2	utput	utput	NOUN
ajst-16801	102	3	z	z	PROPN
ajst-16801	102	4	z	z	PROPN
ajst-16801	102	5	y	y	PROPN
ajst-16801	102	6	y	y	PROPN
ajst-16801	102	7	input	input	NOUN
ajst-16801	102	8	layer	layer	NOUN
ajst-16801	102	9	hidden	hide	VERB
ajst-16801	102	10	layer	layer	NOUN
ajst-16801	102	11	1	1	NUM
ajst-16801	102	12	hidden	hide	VERB
ajst-16801	102	13	layer	layer	NOUN
ajst-16801	102	14	l	l	NOUN
ajst-16801	102	15	output	output	NOUN
ajst-16801	102	16	layer	layer	NOUN
ajst-16801	102	17	238	238	NUM
ajst-16801	102	18	balance	balance	NOUN
ajst-16801	102	19	between	between	ADP
ajst-16801	102	20	the	the	DET
ajst-16801	102	21	dictionary	dictionary	ADJ
ajst-16801	102	22	representation	representation	NOUN
ajst-16801	102	23	and	and	CCONJ
ajst-16801	102	24	the	the	DET
ajst-16801	102	25	discriminant	discriminant	ADJ
ajst-16801	102	26	terms	term	NOUN
ajst-16801	102	27	(	(	PUNCT
ajst-16801	102	28	the	the	DET
ajst-16801	102	29	non	non	ADJ
ajst-16801	102	30	-	-	ADJ
ajst-16801	102	31	convex	convex	ADJ
ajst-16801	102	32	optimization	optimization	NOUN
ajst-16801	102	33	problem	problem	NOUN
ajst-16801	102	34	in	in	ADP
ajst-16801	102	35	(	(	PUNCT
ajst-16801	102	36	4	4	X
ajst-16801	102	37	)	)	PUNCT
ajst-16801	102	38	can	can	AUX
ajst-16801	102	39	be	be	AUX
ajst-16801	102	40	solved	solve	VERB
ajst-16801	102	41	by	by	ADP
ajst-16801	102	42	alternate	alternate	ADJ
ajst-16801	102	43	direction	direction	NOUN
ajst-16801	102	44	method	method	NOUN
ajst-16801	102	45	)	)	PUNCT
ajst-16801	102	46	,	,	PUNCT
ajst-16801	102	47	then	then	ADV
ajst-16801	102	48	,	,	PUNCT
ajst-16801	102	49	the	the	DET
ajst-16801	102	50	network	network	NOUN
ajst-16801	102	51	layer	layer	NOUN
ajst-16801	102	52	shown	show	VERB
ajst-16801	102	53	in	in	ADP
ajst-16801	102	54	figure	figure	NOUN
ajst-16801	102	55	3	3	NUM
ajst-16801	102	56	is	be	AUX
ajst-16801	102	57	used	use	VERB
ajst-16801	102	58	to	to	PART
ajst-16801	102	59	design	design	VERB
ajst-16801	102	60	the	the	DET
ajst-16801	102	61	relevant	relevant	ADJ
ajst-16801	102	62	variables	variable	NOUN
ajst-16801	102	63	,	,	PUNCT
ajst-16801	102	64	and	and	CCONJ
ajst-16801	102	65	finally	finally	ADV
ajst-16801	102	66	,	,	PUNCT
ajst-16801	102	67	the	the	DET
ajst-16801	102	68	bp	bp	PROPN
ajst-16801	102	69	algorithm	algorithm	NOUN
ajst-16801	103	1	[	[	X
ajst-16801	103	2	18	18	NUM
ajst-16801	103	3	]	]	PUNCT
ajst-16801	103	4	is	be	AUX
ajst-16801	103	5	used	use	VERB
ajst-16801	103	6	to	to	PART
ajst-16801	103	7	update	update	VERB
ajst-16801	103	8	the	the	DET
ajst-16801	103	9	network	network	NOUN
ajst-16801	103	10	parameters	parameter	NOUN
ajst-16801	103	11	containing	contain	VERB
ajst-16801	103	12	the	the	DET
ajst-16801	103	13	dictionary	dictionary	ADJ
ajst-16801	103	14	pairs	pair	NOUN
ajst-16801	103	15	.	.	PUNCT
ajst-16801	104	1	for	for	ADP
ajst-16801	104	2	each	each	DET
ajst-16801	104	3	k	k	PROPN
ajst-16801	104	4	class	class	NOUN
ajst-16801	104	5	,	,	PUNCT
ajst-16801	104	6	we	we	PRON
ajst-16801	104	7	can	can	AUX
ajst-16801	104	8	use	use	VERB
ajst-16801	104	9	a	a	DET
ajst-16801	104	10	random	random	ADJ
ajst-16801	104	11	sub	sub	ADJ
ajst-16801	104	12	-	-	ADJ
ajst-16801	104	13	gradient	gradient	ADJ
ajst-16801	104	14	descent	descent	NOUN
ajst-16801	104	15	algorithm	algorithm	NOUN
ajst-16801	104	16	to	to	PART
ajst-16801	104	17	update	update	VERB
ajst-16801	104	18	(	(	PUNCT
ajst-16801	104	19	𝐷	𝐷	NOUN
ajst-16801	104	20	,	,	PUNCT
ajst-16801	104	21	𝑃	𝑃	NOUN
ajst-16801	104	22	)	)	PUNCT
ajst-16801	104	23	)	)	PUNCT
ajst-16801	104	24	on	on	ADP
ajst-16801	104	25	the	the	DET
ajst-16801	104	26	t	t	PROPN
ajst-16801	104	27	iteration	iteration	NOUN
ajst-16801	104	28	until	until	ADP
ajst-16801	104	29	convergence	convergence	NOUN
ajst-16801	104	30	:	:	PUNCT
ajst-16801	104	31	𝐷	𝐷	PROPN
ajst-16801	104	32	𝐷	𝐷	PROPN
ajst-16801	104	33	𝜇	𝜇	ADP
ajst-16801	104	34	𝜕ℒ	𝜕ℒ	PROPN
ajst-16801	104	35	𝜕𝐷	𝜕𝐷	NOUN
ajst-16801	104	36	,	,	PUNCT
ajst-16801	104	37	𝑃	𝑃	NOUN
ajst-16801	104	38	𝑃	𝑃	NOUN
ajst-16801	104	39	𝜇	𝜇	ADP
ajst-16801	104	40	ℒ	ℒ	NOUN
ajst-16801	104	41	,	,	PUNCT
ajst-16801	104	42	(	(	PUNCT
ajst-16801	104	43	5	5	NUM
ajst-16801	104	44	)	)	PUNCT
ajst-16801	104	45	where	where	SCONJ
ajst-16801	104	46	𝜇	𝜇	ADV
ajst-16801	104	47	is	be	AUX
ajst-16801	104	48	the	the	DET
ajst-16801	104	49	learning	learning	NOUN
ajst-16801	104	50	rate	rate	NOUN
ajst-16801	104	51	(	(	PUNCT
ajst-16801	104	52	usually	usually	ADV
ajst-16801	104	53	set	set	VERB
ajst-16801	104	54	to	to	ADP
ajst-16801	104	55	a	a	DET
ajst-16801	104	56	smaller	small	ADJ
ajst-16801	104	57	value	value	NOUN
ajst-16801	104	58	in	in	ADP
ajst-16801	104	59	experiments	experiment	NOUN
ajst-16801	104	60	)	)	PUNCT
ajst-16801	104	61	,	,	PUNCT
ajst-16801	104	62	such	such	ADJ
ajst-16801	104	63	as	as	ADP
ajst-16801	104	64	0.001	0.001	NUM
ajst-16801	104	65	,	,	PUNCT
ajst-16801	104	66	and	and	CCONJ
ajst-16801	104	67	ℒ	ℒ	NOUN
ajst-16801	104	68	relative	relative	ADJ
ajst-16801	104	69	to	to	ADP
ajst-16801	104	70	each	each	DET
ajst-16801	104	71	classspecific	classspecific	ADJ
ajst-16801	104	72	dictionary	dictionary	ADJ
ajst-16801	104	73	pair	pair	NOUN
ajst-16801	104	74	𝐷	𝐷	NOUN
ajst-16801	104	75	and	and	CCONJ
ajst-16801	104	76	𝑃	𝑃	PROPN
ajst-16801	104	77	can	can	AUX
ajst-16801	104	78	be	be	AUX
ajst-16801	104	79	calculated	calculate	VERB
ajst-16801	104	80	as	as	ADP
ajst-16801	104	81	:	:	PUNCT
ajst-16801	104	82	ℒ	ℒ	NOUN
ajst-16801	104	83	𝐷	𝐷	NOUN
ajst-16801	104	84	𝑋	𝑋	NOUN
ajst-16801	104	85	𝑍	𝑍	PROPN
ajst-16801	104	86	𝑋	𝑋	NOUN
ajst-16801	104	87	,	,	PUNCT
ajst-16801	104	88	ℒ	ℒ	NOUN
ajst-16801	104	89	𝜆	𝜆	PRON
ajst-16801	104	90	𝑃	𝑃	NOUN
ajst-16801	104	91	𝑍	𝑍	NOUN
ajst-16801	104	92	𝑋	𝑋	NOUN
ajst-16801	104	93	𝑍	𝑍	VERB
ajst-16801	104	94	𝜆	𝜆	PRON
ajst-16801	104	95	𝑃	𝑃	PROPN
ajst-16801	104	96	�	�	NOUN
ajst-16801	104	97	̅	̅	NOUN
ajst-16801	104	98	�	�	PROPN
ajst-16801	104	99	�	�	PROPN
ajst-16801	104	100	̅	̅	NOUN
ajst-16801	104	101	�	�	PROPN
ajst-16801	104	102	,	,	PUNCT
ajst-16801	104	103	(	(	PUNCT
ajst-16801	104	104	6	6	X
ajst-16801	104	105	)	)	PUNCT
ajst-16801	104	106	the	the	DET
ajst-16801	104	107	gradient	gradient	NOUN
ajst-16801	104	108	in	in	ADP
ajst-16801	104	109	(	(	PUNCT
ajst-16801	104	110	6	6	NUM
ajst-16801	104	111	)	)	PUNCT
ajst-16801	104	112	is	be	AUX
ajst-16801	104	113	then	then	ADV
ajst-16801	104	114	passed	pass	VERB
ajst-16801	104	115	to	to	ADP
ajst-16801	104	116	the	the	DET
ajst-16801	104	117	stack	stack	NOUN
ajst-16801	104	118	autoencoder	autoencoder	NOUN
ajst-16801	104	119	and	and	CCONJ
ajst-16801	104	120	used	use	VERB
ajst-16801	104	121	in	in	ADP
ajst-16801	104	122	standard	standard	ADJ
ajst-16801	104	123	backpropagation	backpropagation	NOUN
ajst-16801	104	124	to	to	PART
ajst-16801	104	125	calculate	calculate	VERB
ajst-16801	104	126	the	the	DET
ajst-16801	104	127	network	network	NOUN
ajst-16801	104	128	parameter	parameter	NOUN
ajst-16801	104	129	gradient	gradient	NOUN
ajst-16801	104	130	𝜕ℒ	𝜕ℒ	PROPN
ajst-16801	104	131	𝜕𝜃	𝜕𝜃	NOUN
ajst-16801	104	132	.	.	PUNCT
ajst-16801	105	1	because	because	SCONJ
ajst-16801	105	2	the	the	DET
ajst-16801	105	3	columns	column	NOUN
ajst-16801	105	4	of	of	ADP
ajst-16801	105	5	both	both	CCONJ
ajst-16801	105	6	the	the	DET
ajst-16801	105	7	analysis	analysis	NOUN
ajst-16801	105	8	dictionary	dictionary	ADJ
ajst-16801	105	9	and	and	CCONJ
ajst-16801	105	10	the	the	DET
ajst-16801	105	11	synthesis	synthesis	NOUN
ajst-16801	105	12	dictionary	dictionary	NOUN
ajst-16801	105	13	are	be	AUX
ajst-16801	105	14	the	the	DET
ajst-16801	105	15	same	same	ADJ
ajst-16801	105	16	as	as	ADP
ajst-16801	105	17	the	the	DET
ajst-16801	105	18	columns	column	NOUN
ajst-16801	105	19	of	of	ADP
ajst-16801	105	20	the	the	DET
ajst-16801	105	21	typical	typical	ADJ
ajst-16801	105	22	class	class	NOUN
ajst-16801	105	23	,	,	PUNCT
ajst-16801	105	24	there	there	PRON
ajst-16801	105	25	is	be	VERB
ajst-16801	105	26	no	no	DET
ajst-16801	105	27	need	need	NOUN
ajst-16801	105	28	to	to	PART
ajst-16801	105	29	update	update	VERB
ajst-16801	105	30	the	the	DET
ajst-16801	105	31	unrelated	unrelated	ADJ
ajst-16801	105	32	subdictionaries	subdictionarie	NOUN
ajst-16801	105	33	represented	represent	VERB
ajst-16801	105	34	by	by	ADP
ajst-16801	105	35	the	the	DET
ajst-16801	105	36	dashed	dash	VERB
ajst-16801	105	37	lines	line	NOUN
ajst-16801	105	38	in	in	ADP
ajst-16801	105	39	figure	figure	NOUN
ajst-16801	105	40	3	3	NUM
ajst-16801	105	41	when	when	SCONJ
ajst-16801	105	42	updating	update	VERB
ajst-16801	105	43	the	the	DET
ajst-16801	105	44	corresponding	corresponding	ADJ
ajst-16801	105	45	dictionary	dictionary	NOUN
ajst-16801	105	46	for	for	ADP
ajst-16801	105	47	class	class	NOUN
ajst-16801	105	48	k.	k.	PROPN
ajst-16801	105	49	figure	figure	NOUN
ajst-16801	105	50	4	4	NUM
ajst-16801	105	51	.	.	PUNCT
ajst-16801	106	1	the	the	DET
ajst-16801	106	2	flow	flow	NOUN
ajst-16801	106	3	chart	chart	NOUN
ajst-16801	106	4	of	of	ADP
ajst-16801	106	5	classification	classification	NOUN
ajst-16801	106	6	of	of	ADP
ajst-16801	106	7	dsdl	dsdl	NOUN
ajst-16801	106	8	as	as	SCONJ
ajst-16801	106	9	shown	show	VERB
ajst-16801	106	10	in	in	ADP
ajst-16801	106	11	figure	figure	NOUN
ajst-16801	106	12	4	4	NUM
ajst-16801	106	13	,	,	PUNCT
ajst-16801	106	14	for	for	ADP
ajst-16801	106	15	such	such	DET
ajst-16801	106	16	a	a	DET
ajst-16801	106	17	complete	complete	ADJ
ajst-16801	106	18	image	image	NOUN
ajst-16801	106	19	classification	classification	NOUN
ajst-16801	106	20	system	system	NOUN
ajst-16801	106	21	,	,	PUNCT
ajst-16801	106	22	it	it	PRON
ajst-16801	106	23	is	be	AUX
ajst-16801	106	24	necessary	necessary	ADJ
ajst-16801	106	25	to	to	PART
ajst-16801	106	26	divide	divide	VERB
ajst-16801	106	27	all	all	DET
ajst-16801	106	28	samples	sample	NOUN
ajst-16801	106	29	in	in	ADP
ajst-16801	106	30	mushroom	mushroom	NOUN
ajst-16801	106	31	image	image	NOUN
ajst-16801	106	32	library	library	NOUN
ajst-16801	106	33	into	into	ADP
ajst-16801	106	34	training	training	NOUN
ajst-16801	106	35	set	set	NOUN
ajst-16801	106	36	and	and	CCONJ
ajst-16801	106	37	test	test	NOUN
ajst-16801	106	38	set	set	VERB
ajst-16801	106	39	,	,	PUNCT
ajst-16801	106	40	use	use	VERB
ajst-16801	106	41	the	the	DET
ajst-16801	106	42	training	training	NOUN
ajst-16801	106	43	set	set	VERB
ajst-16801	106	44	to	to	PART
ajst-16801	106	45	train	train	VERB
ajst-16801	106	46	dsdl	dsdl	PROPN
ajst-16801	106	47	model	model	PROPN
ajst-16801	106	48	,	,	PUNCT
ajst-16801	106	49	obtain	obtain	VERB
ajst-16801	106	50	the	the	DET
ajst-16801	106	51	depth	depth	NOUN
ajst-16801	106	52	sparse	sparse	ADJ
ajst-16801	106	53	features	feature	NOUN
ajst-16801	106	54	and	and	CCONJ
ajst-16801	106	55	dictionary	dictionary	NOUN
ajst-16801	106	56	of	of	ADP
ajst-16801	106	57	the	the	DET
ajst-16801	106	58	training	training	NOUN
ajst-16801	106	59	set	set	NOUN
ajst-16801	106	60	,	,	PUNCT
ajst-16801	106	61	and	and	CCONJ
ajst-16801	106	62	then	then	ADV
ajst-16801	106	63	use	use	VERB
ajst-16801	106	64	the	the	DET
ajst-16801	106	65	dictionary	dictionary	NOUN
ajst-16801	106	66	as	as	ADP
ajst-16801	106	67	the	the	DET
ajst-16801	106	68	network	network	NOUN
ajst-16801	106	69	parameters	parameter	NOUN
ajst-16801	106	70	of	of	ADP
ajst-16801	106	71	a	a	DET
ajst-16801	106	72	new	new	ADJ
ajst-16801	106	73	dsdl	dsdl	NOUN
ajst-16801	106	74	model	model	NOUN
ajst-16801	106	75	and	and	CCONJ
ajst-16801	106	76	import	import	VERB
ajst-16801	106	77	the	the	DET
ajst-16801	106	78	test	test	NOUN
ajst-16801	106	79	set	set	VERB
ajst-16801	106	80	at	at	ADP
ajst-16801	106	81	the	the	DET
ajst-16801	106	82	same	same	ADJ
ajst-16801	106	83	time	time	NOUN
ajst-16801	106	84	to	to	PART
ajst-16801	106	85	obtain	obtain	VERB
ajst-16801	106	86	the	the	DET
ajst-16801	106	87	depth	depth	NOUN
ajst-16801	106	88	features	feature	NOUN
ajst-16801	106	89	of	of	ADP
ajst-16801	106	90	the	the	DET
ajst-16801	106	91	test	test	NOUN
ajst-16801	106	92	set	set	NOUN
ajst-16801	106	93	.	.	PUNCT
ajst-16801	107	1	the	the	DET
ajst-16801	107	2	initialization	initialization	NOUN
ajst-16801	107	3	parameters	parameter	NOUN
ajst-16801	107	4	of	of	ADP
ajst-16801	107	5	the	the	DET
ajst-16801	107	6	dsdl	dsdl	PROPN
ajst-16801	107	7	model	model	NOUN
ajst-16801	107	8	used	use	VERB
ajst-16801	107	9	in	in	ADP
ajst-16801	107	10	the	the	DET
ajst-16801	107	11	training	training	NOUN
ajst-16801	107	12	set	set	NOUN
ajst-16801	107	13	and	and	CCONJ
ajst-16801	107	14	the	the	DET
ajst-16801	107	15	dsdl	dsdl	PROPN
ajst-16801	107	16	model	model	NOUN
ajst-16801	107	17	used	use	VERB
ajst-16801	107	18	in	in	ADP
ajst-16801	107	19	the	the	DET
ajst-16801	107	20	test	test	NOUN
ajst-16801	107	21	set	set	NOUN
ajst-16801	107	22	should	should	AUX
ajst-16801	107	23	be	be	AUX
ajst-16801	107	24	consistent	consistent	ADJ
ajst-16801	107	25	,	,	PUNCT
ajst-16801	107	26	and	and	CCONJ
ajst-16801	107	27	then	then	ADV
ajst-16801	107	28	the	the	DET
ajst-16801	107	29	deep	deep	ADJ
ajst-16801	107	30	sparse	sparse	ADJ
ajst-16801	107	31	features	feature	NOUN
ajst-16801	107	32	of	of	ADP
ajst-16801	107	33	the	the	DET
ajst-16801	107	34	training	training	NOUN
ajst-16801	107	35	set	set	NOUN
ajst-16801	107	36	should	should	AUX
ajst-16801	107	37	be	be	AUX
ajst-16801	107	38	trained	train	VERB
ajst-16801	107	39	on	on	ADP
ajst-16801	107	40	the	the	DET
ajst-16801	107	41	classifier	classifier	NOUN
ajst-16801	107	42	.	.	PUNCT
ajst-16801	108	1	finally	finally	ADV
ajst-16801	108	2	,	,	PUNCT
ajst-16801	108	3	the	the	DET
ajst-16801	108	4	deep	deep	ADJ
ajst-16801	108	5	sparse	sparse	ADJ
ajst-16801	108	6	features	feature	NOUN
ajst-16801	108	7	of	of	ADP
ajst-16801	108	8	the	the	DET
ajst-16801	108	9	test	test	NOUN
ajst-16801	108	10	set	set	NOUN
ajst-16801	108	11	should	should	AUX
ajst-16801	108	12	be	be	AUX
ajst-16801	108	13	input	input	VERB
ajst-16801	108	14	into	into	ADP
ajst-16801	108	15	the	the	DET
ajst-16801	108	16	trained	train	VERB
ajst-16801	108	17	classifier	classifier	NOUN
ajst-16801	108	18	and	and	CCONJ
ajst-16801	108	19	the	the	DET
ajst-16801	108	20	classification	classification	NOUN
ajst-16801	108	21	label	label	NOUN
ajst-16801	108	22	should	should	AUX
ajst-16801	108	23	be	be	AUX
ajst-16801	108	24	predicted	predict	VERB
ajst-16801	108	25	to	to	PART
ajst-16801	108	26	obtain	obtain	VERB
ajst-16801	108	27	the	the	DET
ajst-16801	108	28	final	final	ADJ
ajst-16801	108	29	classification	classification	NOUN
ajst-16801	108	30	result	result	NOUN
ajst-16801	108	31	.	.	PUNCT
ajst-16801	109	1	3	3	X
ajst-16801	109	2	.	.	X
ajst-16801	109	3	analysis	analysis	NOUN
ajst-16801	109	4	and	and	CCONJ
ajst-16801	109	5	discussion	discussion	NOUN
ajst-16801	109	6	algorithm	algorithm	NOUN
ajst-16801	109	7	running	running	NOUN
ajst-16801	109	8	environment	environment	NOUN
ajst-16801	109	9	:	:	PUNCT
ajst-16801	109	10	pycharm2023	pycharm2023	NOUN
ajst-16801	109	11	;	;	PUNCT
ajst-16801	109	12	acceleration	acceleration	NOUN
ajst-16801	109	13	environment	environment	NOUN
ajst-16801	109	14	:	:	PUNCT
ajst-16801	109	15	cuda_11.7	cuda_11.7	NOUN
ajst-16801	109	16	;	;	PUNCT
ajst-16801	109	17	operating	operate	VERB
ajst-16801	109	18	system	system	NOUN
ajst-16801	109	19	:	:	PUNCT
ajst-16801	109	20	windows	window	NOUN
ajst-16801	109	21	10	10	NUM
ajst-16801	109	22	;	;	PUNCT
ajst-16801	109	23	deep	deep	ADJ
ajst-16801	109	24	learning	learning	NOUN
ajst-16801	109	25	framework	framework	NOUN
ajst-16801	109	26	:	:	PUNCT
ajst-16801	109	27	pytorch	pytorch	NOUN
ajst-16801	109	28	2.1.1	2.1.1	NUM
ajst-16801	109	29	;	;	PUNCT
ajst-16801	109	30	language	language	NOUN
ajst-16801	109	31	:	:	PUNCT
ajst-16801	109	32	python	python	PROPN
ajst-16801	109	33	3.7	3.7	NUM
ajst-16801	109	34	.	.	PUNCT
ajst-16801	110	1	3.1	3.1	NUM
ajst-16801	110	2	.	.	PUNCT
ajst-16801	110	3	data	datum	NOUN
ajst-16801	110	4	set	set	VERB
ajst-16801	110	5	description	description	NOUN
ajst-16801	110	6	the	the	DET
ajst-16801	110	7	data	datum	NOUN
ajst-16801	110	8	set	set	VERB
ajst-16801	110	9	used	use	VERB
ajst-16801	110	10	is	be	AUX
ajst-16801	110	11	from	from	ADP
ajst-16801	110	12	the	the	DET
ajst-16801	110	13	kaggle	kaggle	ADJ
ajst-16801	110	14	public	public	ADJ
ajst-16801	110	15	mushroom	mushroom	NOUN
ajst-16801	110	16	dataset	dataset	NOUN
ajst-16801	110	17	,	,	PUNCT
ajst-16801	110	18	which	which	PRON
ajst-16801	110	19	has	have	VERB
ajst-16801	110	20	a	a	DET
ajst-16801	110	21	total	total	NOUN
ajst-16801	110	22	of	of	ADP
ajst-16801	110	23	9533	9533	NUM
ajst-16801	110	24	images	image	NOUN
ajst-16801	110	25	grouped	group	VERB
ajst-16801	110	26	into	into	ADP
ajst-16801	110	27	12	12	NUM
ajst-16801	110	28	categories	category	NOUN
ajst-16801	110	29	.	.	PUNCT
ajst-16801	111	1	the	the	DET
ajst-16801	111	2	data	datum	NOUN
ajst-16801	111	3	set	set	NOUN
ajst-16801	111	4	has	have	VERB
ajst-16801	111	5	two	two	NUM
ajst-16801	111	6	main	main	ADJ
ajst-16801	111	7	characteristics	characteristic	NOUN
ajst-16801	111	8	:	:	PUNCT
ajst-16801	111	9	1	1	X
ajst-16801	111	10	.	.	X
ajst-16801	111	11	there	there	PRON
ajst-16801	111	12	is	be	VERB
ajst-16801	111	13	a	a	DET
ajst-16801	111	14	serious	serious	ADJ
ajst-16801	111	15	imbalance	imbalance	NOUN
ajst-16801	111	16	in	in	ADP
ajst-16801	111	17	the	the	DET
ajst-16801	111	18	number	number	NOUN
ajst-16801	111	19	of	of	ADP
ajst-16801	111	20	samples	sample	NOUN
ajst-16801	111	21	,	,	PUNCT
ajst-16801	111	22	for	for	ADP
ajst-16801	111	23	example	example	NOUN
ajst-16801	111	24	,	,	PUNCT
ajst-16801	111	25	the	the	DET
ajst-16801	111	26	number	number	NOUN
ajst-16801	111	27	of	of	ADP
ajst-16801	111	28	milk	milk	NOUN
ajst-16801	111	29	mushrooms	mushroom	NOUN
ajst-16801	111	30	is	be	AUX
ajst-16801	111	31	as	as	ADV
ajst-16801	111	32	high	high	ADJ
ajst-16801	111	33	as	as	ADP
ajst-16801	111	34	1563	1563	NUM
ajst-16801	111	35	,	,	PUNCT
ajst-16801	111	36	while	while	SCONJ
ajst-16801	111	37	the	the	DET
ajst-16801	111	38	number	number	NOUN
ajst-16801	111	39	of	of	ADP
ajst-16801	111	40	larch	larch	NOUN
ajst-16801	111	41	mushrooms	mushroom	NOUN
ajst-16801	111	42	,	,	PUNCT
ajst-16801	111	43	scarlet	scarlet	ADJ
ajst-16801	111	44	wet	wet	ADJ
ajst-16801	111	45	umbrella	umbrella	NOUN
ajst-16801	111	46	and	and	CCONJ
ajst-16801	111	47	milk	milk	NOUN
ajst-16801	111	48	boletus	boletus	NOUN
ajst-16801	111	49	is	be	AUX
ajst-16801	111	50	less	less	ADJ
ajst-16801	111	51	than	than	ADP
ajst-16801	111	52	400	400	NUM
ajst-16801	111	53	.	.	PUNCT
ajst-16801	112	1	2	2	X
ajst-16801	112	2	.	.	X
ajst-16801	112	3	the	the	DET
ajst-16801	112	4	image	image	NOUN
ajst-16801	112	5	resolution	resolution	NOUN
ajst-16801	112	6	size	size	NOUN
ajst-16801	112	7	is	be	AUX
ajst-16801	112	8	inconsistent	inconsistent	ADJ
ajst-16801	112	9	,	,	PUNCT
ajst-16801	112	10	including	include	VERB
ajst-16801	112	11	multiple	multiple	ADJ
ajst-16801	112	12	resolutions	resolution	NOUN
ajst-16801	112	13	of	of	ADP
ajst-16801	112	14	the	the	DET
ajst-16801	112	15	image	image	NOUN
ajst-16801	112	16	.	.	PUNCT
ajst-16801	113	1	firstly	firstly	ADV
ajst-16801	113	2	,	,	PUNCT
ajst-16801	113	3	the	the	DET
ajst-16801	113	4	image	image	NOUN
ajst-16801	113	5	size	size	NOUN
ajst-16801	113	6	is	be	AUX
ajst-16801	113	7	unified	unify	VERB
ajst-16801	113	8	to	to	ADP
ajst-16801	113	9	300×384	300×384	NUM
ajst-16801	113	10	by	by	ADP
ajst-16801	113	11	preprocessing	preprocesse	VERB
ajst-16801	113	12	,	,	PUNCT
ajst-16801	113	13	and	and	CCONJ
ajst-16801	113	14	then	then	ADV
ajst-16801	113	15	the	the	DET
ajst-16801	113	16	image	image	NOUN
ajst-16801	113	17	is	be	AUX
ajst-16801	113	18	enhanced	enhance	VERB
ajst-16801	113	19	by	by	ADP
ajst-16801	113	20	random	random	ADJ
ajst-16801	113	21	cropping	cropping	NOUN
ajst-16801	113	22	,	,	PUNCT
ajst-16801	113	23	color	color	NOUN
ajst-16801	113	24	distortion	distortion	NOUN
ajst-16801	113	25	,	,	PUNCT
ajst-16801	113	26	image	image	NOUN
ajst-16801	113	27	rotation	rotation	NOUN
ajst-16801	113	28	,	,	PUNCT
ajst-16801	113	29	etc	etc	X
ajst-16801	113	30	.	.	X
ajst-16801	113	31	,	,	PUNCT
ajst-16801	113	32	and	and	CCONJ
ajst-16801	113	33	the	the	DET
ajst-16801	113	34	data	datum	NOUN
ajst-16801	113	35	set	set	VERB
ajst-16801	113	36	is	be	AUX
ajst-16801	113	37	divided	divide	VERB
ajst-16801	113	38	into	into	ADP
ajst-16801	113	39	training	training	NOUN
ajst-16801	113	40	set	set	NOUN
ajst-16801	113	41	,	,	PUNCT
ajst-16801	113	42	verification	verification	NOUN
ajst-16801	113	43	set	set	NOUN
ajst-16801	113	44	and	and	CCONJ
ajst-16801	113	45	test	test	NOUN
ajst-16801	113	46	set	set	VERB
ajst-16801	113	47	according	accord	VERB
ajst-16801	113	48	to	to	ADP
ajst-16801	113	49	the	the	DET
ajst-16801	113	50	ratio	ratio	NOUN
ajst-16801	113	51	of	of	ADP
ajst-16801	113	52	6:2:2	6:2:2	NUM
ajst-16801	113	53	.	.	PUNCT
ajst-16801	114	1	3.2	3.2	NUM
ajst-16801	114	2	.	.	PUNCT
ajst-16801	115	1	training	training	NOUN
ajst-16801	115	2	strategy	strategy	NOUN
ajst-16801	115	3	before	before	ADP
ajst-16801	115	4	training	train	VERB
ajst-16801	115	5	dsdl	dsdl	NOUN
ajst-16801	115	6	algorithm	algorithm	PROPN
ajst-16801	115	7	,	,	PUNCT
ajst-16801	115	8	proper	proper	ADJ
ajst-16801	115	9	pre	pre	ADJ
ajst-16801	115	10	-	-	ADJ
ajst-16801	115	11	training	training	ADJ
ajst-16801	115	12	and	and	CCONJ
ajst-16801	115	13	fine	fine	ADV
ajst-16801	115	14	-	-	PUNCT
ajst-16801	115	15	tuning	tuning	NOUN
ajst-16801	115	16	strategies	strategy	NOUN
ajst-16801	115	17	can	can	AUX
ajst-16801	115	18	minimize	minimize	VERB
ajst-16801	115	19	the	the	DET
ajst-16801	115	20	reconstruction	reconstruction	NOUN
ajst-16801	115	21	error	error	NOUN
ajst-16801	115	22	between	between	ADP
ajst-16801	115	23	y	y	PROPN
ajst-16801	115	24	and	and	CCONJ
ajst-16801	115	25	y	y	PROPN
ajst-16801	115	26	̂	̂	PUNCT
ajst-16801	115	27	without	without	ADP
ajst-16801	115	28	getting	get	VERB
ajst-16801	115	29	a	a	DET
ajst-16801	115	30	trivial	trivial	ADJ
ajst-16801	115	31	all	all	ADV
ajst-16801	115	32	-	-	PUNCT
ajst-16801	115	33	zero	zero	NUM
ajst-16801	115	34	solution	solution	NOUN
ajst-16801	115	35	.	.	PUNCT
ajst-16801	116	1	we	we	PRON
ajst-16801	116	2	only	only	ADV
ajst-16801	116	3	use	use	VERB
ajst-16801	116	4	all	all	DET
ajst-16801	116	5	the	the	DET
ajst-16801	116	6	training	training	NOUN
ajst-16801	116	7	images	image	NOUN
ajst-16801	116	8	to	to	PART
ajst-16801	116	9	train	train	VERB
ajst-16801	116	10	the	the	DET
ajst-16801	116	11	encoder	encoder	NOUN
ajst-16801	116	12	and	and	CCONJ
ajst-16801	116	13	decoder	decoder	NOUN
ajst-16801	116	14	.	.	PUNCT
ajst-16801	117	1	the	the	DET
ajst-16801	117	2	encoder	encoder	NOUN
ajst-16801	117	3	and	and	CCONJ
ajst-16801	117	4	decoder	decoder	NOUN
ajst-16801	117	5	are	be	AUX
ajst-16801	117	6	then	then	ADV
ajst-16801	117	7	initialized	initialize	VERB
ajst-16801	117	8	with	with	ADP
ajst-16801	117	9	the	the	DET
ajst-16801	117	10	updated	update	VERB
ajst-16801	117	11	parameters	parameter	NOUN
ajst-16801	117	12	.	.	PUNCT
ajst-16801	118	1	in	in	ADP
ajst-16801	118	2	the	the	DET
ajst-16801	118	3	fine	fine	ADV
ajst-16801	118	4	-	-	PUNCT
ajst-16801	118	5	tuning	tune	VERB
ajst-16801	118	6	stage	stage	NOUN
ajst-16801	118	7	,	,	PUNCT
ajst-16801	118	8	the	the	DET
ajst-16801	118	9	gradient	gradient	ADJ
ajst-16801	118	10	descent	descent	NOUN
ajst-16801	118	11	method	method	NOUN
ajst-16801	118	12	(	(	PUNCT
ajst-16801	118	13	4	4	NUM
ajst-16801	118	14	)	)	PUNCT
ajst-16801	118	15	is	be	AUX
ajst-16801	118	16	applied	apply	VERB
ajst-16801	118	17	.	.	PUNCT
ajst-16801	119	1	after	after	SCONJ
ajst-16801	119	2	the	the	DET
ajst-16801	119	3	whole	whole	ADJ
ajst-16801	119	4	network	network	NOUN
ajst-16801	119	5	is	be	AUX
ajst-16801	119	6	trained	train	VERB
ajst-16801	119	7	,	,	PUNCT
ajst-16801	119	8	the	the	DET
ajst-16801	119	9	reconstruction	reconstruction	NOUN
ajst-16801	119	10	error	error	NOUN
ajst-16801	119	11	of	of	ADP
ajst-16801	119	12	image	image	NOUN
ajst-16801	119	13	classification	classification	NOUN
ajst-16801	119	14	is	be	AUX
ajst-16801	119	15	calculated	calculate	VERB
ajst-16801	119	16	using	use	VERB
ajst-16801	119	17	the	the	DET
ajst-16801	119	18	parameters	parameter	NOUN
ajst-16801	119	19	of	of	ADP
ajst-16801	119	20	the	the	DET
ajst-16801	119	21	encoder	encoder	NOUN
ajst-16801	119	22	layer	layer	NOUN
ajst-16801	119	23	and	and	CCONJ
ajst-16801	119	24	the	the	DET
ajst-16801	119	25	dictionary	dictionary	ADJ
ajst-16801	119	26	pair	pair	NOUN
ajst-16801	119	27	:	:	PUNCT
ajst-16801	119	28	label(y)=argmin	label(y)=argmin	ADJ
ajst-16801	119	29	∥	∥	NUM
ajst-16801	119	30	h	h	NOUN
ajst-16801	119	31	y	y	NOUN
ajst-16801	119	32	;	;	PUNCT
ajst-16801	119	33	𝜃	𝜃	NUM
ajst-16801	119	34	𝐷	𝐷	NOUN
ajst-16801	119	35	𝑃	𝑃	VERB
ajst-16801	119	36	ℎ	ℎ	NOUN
ajst-16801	119	37	𝑦	𝑦	NOUN
ajst-16801	119	38	;	;	PUNCT
ajst-16801	119	39	𝜃	𝜃	X
ajst-16801	119	40	∥	∥	X
ajst-16801	119	41	（	（	PUNCT
ajst-16801	119	42	7	7	NUM
ajst-16801	119	43	）	）	SYM
ajst-16801	119	44	where	where	SCONJ
ajst-16801	119	45	y∈	y∈	PROPN
ajst-16801	119	46	𝑅	𝑅	PROPN
ajst-16801	119	47	is	be	AUX
ajst-16801	119	48	a	a	DET
ajst-16801	119	49	vectorized	vectorize	VERB
ajst-16801	119	50	test	test	NOUN
ajst-16801	119	51	image	image	NOUN
ajst-16801	119	52	,	,	PUNCT
ajst-16801	119	53	as	as	SCONJ
ajst-16801	119	54	shown	show	VERB
ajst-16801	119	55	in	in	ADP
ajst-16801	119	56	table	table	NOUN
ajst-16801	119	57	1.1	1.1	NUM
ajst-16801	119	58	,	,	PUNCT
ajst-16801	119	59	the	the	DET
ajst-16801	119	60	steps	step	NOUN
ajst-16801	119	61	of	of	ADP
ajst-16801	119	62	mushroom	mushroom	NOUN
ajst-16801	119	63	image	image	NOUN
ajst-16801	119	64	classification	classification	NOUN
ajst-16801	119	65	by	by	ADP
ajst-16801	119	66	dsdl	dsdl	PROPN
ajst-16801	119	67	algorithm	algorithm	PROPN
ajst-16801	119	68	are	be	AUX
ajst-16801	119	69	as	as	SCONJ
ajst-16801	119	70	follows	follow	VERB
ajst-16801	119	71	:	:	PUNCT
ajst-16801	119	72	table	table	NOUN
ajst-16801	119	73	1	1	NUM
ajst-16801	119	74	.	.	PUNCT
ajst-16801	120	1	the	the	DET
ajst-16801	120	2	step	step	NOUN
ajst-16801	120	3	of	of	ADP
ajst-16801	120	4	dsdl	dsdl	PROPN
ajst-16801	120	5	dsdl	dsdl	PROPN
ajst-16801	120	6	algorithm	algorithm	PROPN
ajst-16801	120	7	flow	flow	NOUN
ajst-16801	120	8	:	:	PUNCT
ajst-16801	120	9	a	a	X
ajst-16801	120	10	)	)	PUNCT
ajst-16801	120	11	input	input	NOUN
ajst-16801	120	12	:	:	PUNCT
ajst-16801	120	13	training	train	VERB
ajst-16801	120	14	image	image	NOUN
ajst-16801	120	15	y∈	y∈	PROPN
ajst-16801	120	16	𝑅	𝑅	PROPN
ajst-16801	120	17	,	,	PUNCT
ajst-16801	120	18	the	the	DET
ajst-16801	120	19	atomic	atomic	ADJ
ajst-16801	120	20	number	number	NOUN
ajst-16801	120	21	of	of	ADP
ajst-16801	120	22	each	each	DET
ajst-16801	120	23	class	class	NOUN
ajst-16801	120	24	d	d	PROPN
ajst-16801	120	25	,	,	PUNCT
ajst-16801	120	26	parameter{𝜆	parameter{𝜆	PROPN
ajst-16801	120	27	,	,	PUNCT
ajst-16801	120	28	𝜆	𝜆	X
ajst-16801	120	29	}	}	PUNCT
ajst-16801	120	30	,	,	PUNCT
ajst-16801	120	31	test	test	NOUN
ajst-16801	120	32	image	image	NOUN
ajst-16801	120	33	y∈	y∈	PROPN
ajst-16801	120	34	𝑅	𝑅	PROPN
ajst-16801	120	35	.	.	PUNCT
ajst-16801	120	36	b	b	X
ajst-16801	120	37	)	)	PUNCT
ajst-16801	120	38	by	by	ADP
ajst-16801	120	39	training	train	VERB
ajst-16801	120	40	the	the	DET
ajst-16801	120	41	autoencoder	autoencoder	NOUN
ajst-16801	120	42	network	network	NOUN
ajst-16801	120	43	,	,	PUNCT
ajst-16801	120	44	initialize	initialize	VERB
ajst-16801	120	45	the	the	DET
ajst-16801	120	46	whole	whole	ADJ
ajst-16801	120	47	network	network	NOUN
ajst-16801	120	48	with	with	ADP
ajst-16801	120	49	the	the	DET
ajst-16801	120	50	pre	pre	ADJ
ajst-16801	120	51	-	-	ADJ
ajst-16801	120	52	trained	trained	ADJ
ajst-16801	120	53	parameters	parameter	NOUN
ajst-16801	120	54	.	.	PUNCT
ajst-16801	121	1	c	c	X
ajst-16801	121	2	)	)	PUNCT
ajst-16801	121	3	update	update	NOUN
ajst-16801	121	4	dictionary	dictionary	ADJ
ajst-16801	121	5	and	and	CCONJ
ajst-16801	121	6	network	network	NOUN
ajst-16801	121	7	parameters	parameter	NOUN
ajst-16801	121	8	by	by	ADP
ajst-16801	121	9	bp	bp	PROPN
ajst-16801	121	10	algorithm	algorithm	PROPN
ajst-16801	121	11	.	.	PUNCT
ajst-16801	122	1	d	d	X
ajst-16801	122	2	)	)	PUNCT
ajst-16801	122	3	using	use	VERB
ajst-16801	122	4	the	the	DET
ajst-16801	122	5	dictionary	dictionary	NOUN
ajst-16801	122	6	for	for	ADP
ajst-16801	122	7	{	{	PUNCT
ajst-16801	122	8	d	d	PROPN
ajst-16801	122	9	,	,	PUNCT
ajst-16801	122	10	p	p	NOUN
ajst-16801	122	11	}	}	PUNCT
ajst-16801	122	12	and	and	CCONJ
ajst-16801	122	13	the	the	DET
ajst-16801	122	14	encoder	encoder	NOUN
ajst-16801	122	15	parameter	parameter	PROPN
ajst-16801	122	16	𝜃	𝜃	PROPN
ajst-16801	122	17	,	,	PUNCT
ajst-16801	122	18	the	the	DET
ajst-16801	122	19	k	k	PROPN
ajst-16801	122	20	reconstruction	reconstruction	NOUN
ajst-16801	122	21	error	error	NOUN
ajst-16801	122	22	is	be	AUX
ajst-16801	122	23	calculated	calculate	VERB
ajst-16801	122	24	by	by	ADP
ajst-16801	122	25	239	239	NUM
ajst-16801	122	26	formula	formula	NOUN
ajst-16801	122	27	(	(	PUNCT
ajst-16801	122	28	7	7	NUM
ajst-16801	122	29	)	)	PUNCT
ajst-16801	122	30	.	.	PUNCT
ajst-16801	123	1	e	e	X
ajst-16801	123	2	)	)	PUNCT
ajst-16801	123	3	implement	implement	VERB
ajst-16801	123	4	the	the	DET
ajst-16801	123	5	result	result	NOUN
ajst-16801	123	6	label	label	NOUN
ajst-16801	123	7	of	of	ADP
ajst-16801	123	8	the	the	DET
ajst-16801	123	9	test	test	NOUN
ajst-16801	123	10	image	image	NOUN
ajst-16801	123	11	through	through	ADP
ajst-16801	123	12	(	(	PUNCT
ajst-16801	123	13	7	7	NUM
ajst-16801	123	14	)	)	PUNCT
ajst-16801	123	15	.	.	PUNCT
ajst-16801	124	1	f	f	X
ajst-16801	124	2	)	)	PUNCT
ajst-16801	124	3	output	output	NOUN
ajst-16801	124	4	:	:	PUNCT
ajst-16801	124	5	label	label	NOUN
ajst-16801	124	6	of	of	ADP
ajst-16801	124	7	the	the	DET
ajst-16801	124	8	test	test	NOUN
ajst-16801	124	9	image	image	NOUN
ajst-16801	124	10	:	:	PUNCT
ajst-16801	124	11	label	label	NOUN
ajst-16801	124	12	(	(	PUNCT
ajst-16801	124	13	y	y	NOUN
ajst-16801	124	14	)	)	PUNCT
ajst-16801	124	15	.	.	PUNCT
ajst-16801	125	1	in	in	ADP
ajst-16801	125	2	table	table	NOUN
ajst-16801	125	3	1	1	NUM
ajst-16801	125	4	,	,	PUNCT
ajst-16801	125	5	it	it	PRON
ajst-16801	125	6	can	can	AUX
ajst-16801	125	7	be	be	AUX
ajst-16801	125	8	observed	observe	VERB
ajst-16801	125	9	that	that	SCONJ
ajst-16801	125	10	all	all	DET
ajst-16801	125	11	the	the	DET
ajst-16801	125	12	training	training	NOUN
ajst-16801	125	13	images	image	NOUN
ajst-16801	125	14	share	share	VERB
ajst-16801	125	15	an	an	DET
ajst-16801	125	16	autoencoder	autoencoder	NOUN
ajst-16801	125	17	network	network	NOUN
ajst-16801	125	18	that	that	PRON
ajst-16801	125	19	generates	generate	VERB
ajst-16801	125	20	the	the	DET
ajst-16801	125	21	depth	depth	NOUN
ajst-16801	125	22	features	feature	NOUN
ajst-16801	125	23	used	use	VERB
ajst-16801	125	24	for	for	ADP
ajst-16801	125	25	dictionary	dictionary	ADJ
ajst-16801	125	26	pair	pair	NOUN
ajst-16801	125	27	learning	learning	NOUN
ajst-16801	125	28	.	.	PUNCT
ajst-16801	126	1	in	in	ADP
ajst-16801	126	2	addition	addition	NOUN
ajst-16801	126	3	,	,	PUNCT
ajst-16801	126	4	structured	structured	ADJ
ajst-16801	126	5	dictionary	dictionary	ADJ
ajst-16801	126	6	pairs	pair	NOUN
ajst-16801	126	7	are	be	AUX
ajst-16801	126	8	trained	train	VERB
ajst-16801	126	9	to	to	PART
ajst-16801	126	10	represent	represent	VERB
ajst-16801	126	11	a	a	DET
ajst-16801	126	12	specific	specific	ADJ
ajst-16801	126	13	class	class	NOUN
ajst-16801	126	14	of	of	ADP
ajst-16801	126	15	trained	train	VERB
ajst-16801	126	16	images	image	NOUN
ajst-16801	126	17	to	to	PART
ajst-16801	126	18	aid	aid	VERB
ajst-16801	126	19	in	in	ADP
ajst-16801	126	20	image	image	NOUN
ajst-16801	126	21	classification	classification	NOUN
ajst-16801	126	22	.	.	PUNCT
ajst-16801	127	1	the	the	DET
ajst-16801	127	2	stack	stack	NOUN
ajst-16801	127	3	autoencoder	autoencoder	NOUN
ajst-16801	127	4	is	be	AUX
ajst-16801	127	5	introduced	introduce	VERB
ajst-16801	127	6	into	into	ADP
ajst-16801	127	7	the	the	DET
ajst-16801	127	8	dictionary	dictionary	ADJ
ajst-16801	127	9	pair	pair	NOUN
ajst-16801	127	10	learning	learn	VERB
ajst-16801	127	11	framework	framework	NOUN
ajst-16801	127	12	,	,	PUNCT
ajst-16801	127	13	which	which	PRON
ajst-16801	127	14	bridges	bridge	VERB
ajst-16801	127	15	the	the	DET
ajst-16801	127	16	gap	gap	NOUN
ajst-16801	127	17	between	between	ADP
ajst-16801	127	18	deep	deep	ADJ
ajst-16801	127	19	feature	feature	NOUN
ajst-16801	127	20	representation	representation	NOUN
ajst-16801	127	21	and	and	CCONJ
ajst-16801	127	22	discriminative	discriminative	NOUN
ajst-16801	127	23	dictionary	dictionary	ADJ
ajst-16801	127	24	learning	learning	NOUN
ajst-16801	127	25	.	.	PUNCT
ajst-16801	128	1	dsdl	dsdl	PROPN
ajst-16801	128	2	introduces	introduce	VERB
ajst-16801	128	3	a	a	DET
ajst-16801	128	4	common	common	ADJ
ajst-16801	128	5	unsupervised	unsupervised	ADJ
ajst-16801	128	6	autoencoder	autoencoder	NOUN
ajst-16801	128	7	architecture	architecture	NOUN
ajst-16801	128	8	that	that	PRON
ajst-16801	128	9	does	do	AUX
ajst-16801	128	10	not	not	PART
ajst-16801	128	11	require	require	VERB
ajst-16801	128	12	a	a	DET
ajst-16801	128	13	third	third	ADJ
ajst-16801	128	14	-	-	PUNCT
ajst-16801	128	15	party	party	NOUN
ajst-16801	128	16	pre	pre	ADJ
ajst-16801	128	17	-	-	ADJ
ajst-16801	128	18	trained	train	VERB
ajst-16801	128	19	model	model	NOUN
ajst-16801	128	20	,	,	PUNCT
ajst-16801	128	21	such	such	ADJ
ajst-16801	128	22	as	as	ADP
ajst-16801	128	23	resnet	resnet	NOUN
ajst-16801	128	24	.	.	PUNCT
ajst-16801	129	1	only	only	ADV
ajst-16801	129	2	a	a	DET
ajst-16801	129	3	limited	limited	ADJ
ajst-16801	129	4	number	number	NOUN
ajst-16801	129	5	of	of	ADP
ajst-16801	129	6	layers	layer	NOUN
ajst-16801	129	7	are	be	AUX
ajst-16801	129	8	used	use	VERB
ajst-16801	129	9	to	to	PART
ajst-16801	129	10	explore	explore	VERB
ajst-16801	129	11	in	in	ADP
ajst-16801	129	12	an	an	DET
ajst-16801	129	13	unsupervised	unsupervised	ADJ
ajst-16801	129	14	manner	manner	NOUN
ajst-16801	129	15	,	,	PUNCT
ajst-16801	129	16	resulting	result	VERB
ajst-16801	129	17	in	in	ADP
ajst-16801	129	18	features	feature	NOUN
ajst-16801	129	19	with	with	ADP
ajst-16801	129	20	better	well	ADJ
ajst-16801	129	21	depth	depth	NOUN
ajst-16801	129	22	than	than	ADP
ajst-16801	129	23	the	the	DET
ajst-16801	129	24	original	original	ADJ
ajst-16801	129	25	features	feature	NOUN
ajst-16801	129	26	,	,	PUNCT
ajst-16801	129	27	so	so	SCONJ
ajst-16801	129	28	that	that	SCONJ
ajst-16801	129	29	the	the	DET
ajst-16801	129	30	proposed	propose	VERB
ajst-16801	129	31	deep	deep	ADJ
ajst-16801	129	32	sparse	sparse	ADJ
ajst-16801	129	33	dictionary	dictionary	ADJ
ajst-16801	129	34	learning	learning	NOUN
ajst-16801	129	35	can	can	AUX
ajst-16801	129	36	be	be	AUX
ajst-16801	129	37	carried	carry	VERB
ajst-16801	129	38	out	out	ADP
ajst-16801	129	39	in	in	ADP
ajst-16801	129	40	the	the	DET
ajst-16801	129	41	potential	potential	ADJ
ajst-16801	129	42	feature	feature	NOUN
ajst-16801	129	43	space	space	NOUN
ajst-16801	129	44	and	and	CCONJ
ajst-16801	129	45	show	show	VERB
ajst-16801	129	46	its	its	PRON
ajst-16801	129	47	performance	performance	NOUN
ajst-16801	129	48	.	.	PUNCT
ajst-16801	130	1	3.3	3.3	NUM
ajst-16801	130	2	.	.	PUNCT
ajst-16801	131	1	experimental	experimental	ADJ
ajst-16801	131	2	results	result	NOUN
ajst-16801	131	3	and	and	CCONJ
ajst-16801	131	4	analysis	analysis	NOUN
ajst-16801	131	5	we	we	PRON
ajst-16801	131	6	use	use	VERB
ajst-16801	131	7	a	a	DET
ajst-16801	131	8	publicly	publicly	ADV
ajst-16801	131	9	available	available	ADJ
ajst-16801	131	10	mushroom	mushroom	NOUN
ajst-16801	131	11	image	image	NOUN
ajst-16801	131	12	dataset	dataset	VERB
ajst-16801	131	13	to	to	PART
ajst-16801	131	14	evaluate	evaluate	VERB
ajst-16801	131	15	the	the	DET
ajst-16801	131	16	proposed	propose	VERB
ajst-16801	131	17	dsdl	dsdl	NOUN
ajst-16801	131	18	image	image	NOUN
ajst-16801	131	19	classification	classification	NOUN
ajst-16801	131	20	algorithm	algorithm	NOUN
ajst-16801	131	21	.	.	PUNCT
ajst-16801	132	1	the	the	DET
ajst-16801	132	2	proposed	propose	VERB
ajst-16801	132	3	algorithm	algorithm	NOUN
ajst-16801	132	4	is	be	AUX
ajst-16801	132	5	compared	compare	VERB
ajst-16801	132	6	with	with	ADP
ajst-16801	132	7	other	other	ADJ
ajst-16801	132	8	image	image	NOUN
ajst-16801	132	9	classification	classification	NOUN
ajst-16801	132	10	methods	method	NOUN
ajst-16801	132	11	of	of	ADP
ajst-16801	132	12	dictionary	dictionary	ADJ
ajst-16801	132	13	learning	learning	NOUN
ajst-16801	132	14	.	.	PUNCT
ajst-16801	133	1	that	that	PRON
ajst-16801	133	2	is	be	AUX
ajst-16801	133	3	,	,	PUNCT
ajst-16801	133	4	classification	classification	NOUN
ajst-16801	133	5	based	base	VERB
ajst-16801	133	6	on	on	ADP
ajst-16801	133	7	sparse	sparse	ADJ
ajst-16801	133	8	representation	representation	NOUN
ajst-16801	133	9	(	(	PUNCT
ajst-16801	133	10	src)[19	src)[19	NOUN
ajst-16801	133	11	]	]	X
ajst-16801	133	12	,	,	PUNCT
ajst-16801	133	13	classification	classification	NOUN
ajst-16801	133	14	based	base	VERB
ajst-16801	133	15	on	on	ADP
ajst-16801	133	16	block	block	NOUN
ajst-16801	133	17	sparse	sparse	ADJ
ajst-16801	133	18	representation	representation	NOUN
ajst-16801	133	19	(	(	PUNCT
ajst-16801	133	20	bsrc)[20	bsrc)[20	PROPN
ajst-16801	133	21	]	]	PUNCT
ajst-16801	133	22	(	(	PUNCT
ajst-16801	133	23	in	in	ADP
ajst-16801	133	24	which	which	PRON
ajst-16801	133	25	the	the	DET
ajst-16801	133	26	block	block	NOUN
ajst-16801	133	27	sparse	sparse	ADJ
ajst-16801	133	28	coding	code	VERB
ajst-16801	133	29	problem	problem	NOUN
ajst-16801	133	30	uses	use	VERB
ajst-16801	133	31	the	the	DET
ajst-16801	133	32	cvx	cvx	PROPN
ajst-16801	133	33	toolbox	toolbox	NOUN
ajst-16801	133	34	[	[	X
ajst-16801	133	35	21	21	NUM
ajst-16801	133	36	]	]	PUNCT
ajst-16801	133	37	)	)	PUNCT
ajst-16801	133	38	,	,	PUNCT
ajst-16801	133	39	classification	classification	NOUN
ajst-16801	133	40	based	base	VERB
ajst-16801	133	41	on	on	ADP
ajst-16801	133	42	cooperative	cooperative	ADJ
ajst-16801	133	43	representation	representation	NOUN
ajst-16801	133	44	(	(	PUNCT
ajst-16801	133	45	crc)[22	crc)[22	NOUN
ajst-16801	133	46	]	]	X
ajst-16801	133	47	,	,	PUNCT
ajst-16801	133	48	and	and	CCONJ
ajst-16801	133	49	lc	lc	NOUN
ajst-16801	133	50	-	-	PUNCT
ajst-16801	133	51	ksvd	ksvd	NOUN
ajst-16801	133	52	(	(	PUNCT
ajst-16801	133	53	lc	lc	NOUN
ajst-16801	133	54	-	-	PUNCT
ajst-16801	133	55	ksvd2	ksvd2	NOUN
ajst-16801	133	56	)	)	PUNCT
ajst-16801	133	57	in[23	in[23	PROPN
ajst-16801	133	58	]	]	X
ajst-16801	133	59	)	)	PUNCT
ajst-16801	133	60	,	,	PUNCT
ajst-16801	133	61	fddl[24	fddl[24	PROPN
ajst-16801	133	62	]	]	PUNCT
ajst-16801	133	63	,	,	PUNCT
ajst-16801	133	64	support	support	NOUN
ajst-16801	133	65	vector	vector	NOUN
ajst-16801	133	66	guided	guide	VERB
ajst-16801	133	67	dictionary	dictionary	ADJ
ajst-16801	133	68	learning	learning	NOUN
ajst-16801	133	69	(	(	PUNCT
ajst-16801	133	70	svgdl)[25	svgdl)[25	PROPN
ajst-16801	133	71	]	]	PUNCT
ajst-16801	133	72	,	,	PUNCT
ajst-16801	133	73	dictionary	dictionary	ADJ
ajst-16801	133	74	pair	pair	NOUN
ajst-16801	133	75	learning	learning	NOUN
ajst-16801	133	76	(	(	PUNCT
ajst-16801	133	77	dpl)[26	dpl)[26	PROPN
ajst-16801	133	78	]	]	PUNCT
ajst-16801	133	79	,	,	PUNCT
ajst-16801	133	80	efficient	efficient	ADJ
ajst-16801	133	81	and	and	CCONJ
ajst-16801	133	82	robust	robust	ADJ
ajst-16801	133	83	discriminant	discriminant	NOUN
ajst-16801	133	84	dictionary	dictionary	ADJ
ajst-16801	133	85	pair	pair	NOUN
ajst-16801	133	86	learning	learning	NOUN
ajst-16801	133	87	(	(	PUNCT
ajst-16801	133	88	erddpl)[27	erddpl)[27	PROPN
ajst-16801	133	89	]	]	PUNCT
ajst-16801	133	90	,	,	PUNCT
ajst-16801	133	91	slatdpl[28	slatdpl[28	PROPN
ajst-16801	133	92	]	]	PUNCT
ajst-16801	133	93	,	,	PUNCT
ajst-16801	133	94	and	and	CCONJ
ajst-16801	133	95	support	support	VERB
ajst-16801	133	96	vector	vector	NOUN
ajst-16801	133	97	machine	machine	NOUN
ajst-16801	133	98	embedded	embed	VERB
ajst-16801	133	99	discriminant	discriminant	ADJ
ajst-16801	133	100	dictionary	dictionary	ADJ
ajst-16801	133	101	pair	pair	NOUN
ajst-16801	133	102	learning	learning	NOUN
ajst-16801	133	103	(	(	PUNCT
ajst-16801	133	104	svmddpl)[29	svmddpl)[29	NOUN
ajst-16801	133	105	]	]	PUNCT
ajst-16801	133	106	.	.	PUNCT
ajst-16801	134	1	in	in	ADP
ajst-16801	134	2	the	the	DET
ajst-16801	134	3	experimental	experimental	ADJ
ajst-16801	134	4	comparison	comparison	NOUN
ajst-16801	134	5	,	,	PUNCT
ajst-16801	134	6	we	we	PRON
ajst-16801	134	7	trained	train	VERB
ajst-16801	134	8	the	the	DET
ajst-16801	134	9	correlation	correlation	NOUN
ajst-16801	134	10	algorithm	algorithm	NOUN
ajst-16801	134	11	with	with	ADP
ajst-16801	134	12	20	20	NUM
ajst-16801	134	13	iterations	iteration	NOUN
ajst-16801	134	14	respectively	respectively	ADV
ajst-16801	134	15	,	,	PUNCT
ajst-16801	134	16	updated	update	VERB
ajst-16801	134	17	the	the	DET
ajst-16801	134	18	learning	learning	NOUN
ajst-16801	134	19	rate	rate	NOUN
ajst-16801	134	20	𝜂	𝜂	NOUN
ajst-16801	134	21	=	=	SYM
ajst-16801	134	22	10	10	NUM
ajst-16801	134	23	of	of	ADP
ajst-16801	134	24	the	the	DET
ajst-16801	134	25	correlation	correlation	NOUN
ajst-16801	134	26	variable	variable	NOUN
ajst-16801	134	27	,	,	PUNCT
ajst-16801	134	28	and	and	CCONJ
ajst-16801	134	29	took	take	VERB
ajst-16801	134	30	the	the	DET
ajst-16801	134	31	average	average	ADJ
ajst-16801	134	32	value	value	NOUN
ajst-16801	134	33	of	of	ADP
ajst-16801	134	34	the	the	DET
ajst-16801	134	35	classification	classification	NOUN
ajst-16801	134	36	results	result	NOUN
ajst-16801	134	37	for	for	ADP
ajst-16801	134	38	10	10	NUM
ajst-16801	134	39	runs	run	NOUN
ajst-16801	134	40	.	.	PUNCT
ajst-16801	135	1	parameters	parameter	NOUN
ajst-16801	135	2	𝜆	𝜆	SCONJ
ajst-16801	135	3	𝜆	𝜆	X
ajst-16801	135	4	from	from	ADP
ajst-16801	135	5	{	{	PUNCT
ajst-16801	135	6	0.01	0.01	NUM
ajst-16801	135	7	,	,	PUNCT
ajst-16801	135	8	0.02	0.02	NUM
ajst-16801	135	9	,	,	PUNCT
ajst-16801	135	10	0.03	0.03	NUM
ajst-16801	135	11	,	,	PUNCT
ajst-16801	135	12	0.04	0.04	NUM
ajst-16801	135	13	,	,	PUNCT
ajst-16801	135	14	0.05	0.05	NUM
ajst-16801	135	15	}	}	PUNCT
ajst-16801	135	16	{	{	PUNCT
ajst-16801	135	17	0.001	0.001	NUM
ajst-16801	135	18	,	,	PUNCT
ajst-16801	135	19	0.005	0.005	NUM
ajst-16801	135	20	,	,	PUNCT
ajst-16801	135	21	0.01	0.01	NUM
ajst-16801	135	22	,	,	PUNCT
ajst-16801	135	23	0.05	0.05	NUM
ajst-16801	135	24	}	}	PUNCT
ajst-16801	135	25	grid	grid	NOUN
ajst-16801	135	26	.	.	PUNCT
ajst-16801	136	1	in	in	ADP
ajst-16801	136	2	the	the	DET
ajst-16801	136	3	experiment	experiment	NOUN
ajst-16801	136	4	,	,	PUNCT
ajst-16801	136	5	relu	relu	NOUN
ajst-16801	136	6	was	be	AUX
ajst-16801	136	7	used	use	VERB
ajst-16801	136	8	as	as	ADP
ajst-16801	136	9	the	the	DET
ajst-16801	136	10	activation	activation	NOUN
ajst-16801	136	11	function	function	NOUN
ajst-16801	136	12	,	,	PUNCT
ajst-16801	136	13	and	and	CCONJ
ajst-16801	136	14	the	the	DET
ajst-16801	136	15	parameters	parameter	NOUN
ajst-16801	136	16	of	of	ADP
ajst-16801	136	17	the	the	DET
ajst-16801	136	18	autoencoder	autoencoder	NOUN
ajst-16801	136	19	,	,	PUNCT
ajst-16801	136	20	such	such	ADJ
ajst-16801	136	21	as	as	ADP
ajst-16801	136	22	the	the	DET
ajst-16801	136	23	convolution	convolution	NOUN
ajst-16801	136	24	kernel	kernel	PROPN
ajst-16801	136	25	size	size	NOUN
ajst-16801	136	26	mentioned	mention	VERB
ajst-16801	136	27	in	in	ADP
ajst-16801	136	28	[	[	X
ajst-16801	136	29	30,31	30,31	PROPN
ajst-16801	136	30	]	]	PUNCT
ajst-16801	136	31	,	,	PUNCT
ajst-16801	136	32	were	be	AUX
ajst-16801	136	33	first	first	ADV
ajst-16801	136	34	used	use	VERB
ajst-16801	136	35	,	,	PUNCT
ajst-16801	136	36	and	and	CCONJ
ajst-16801	136	37	then	then	ADV
ajst-16801	136	38	fine	fine	ADV
ajst-16801	136	39	-	-	PUNCT
ajst-16801	136	40	tuned	tune	VERB
ajst-16801	136	41	on	on	ADP
ajst-16801	136	42	the	the	DET
ajst-16801	136	43	mushroom	mushroom	NOUN
ajst-16801	136	44	dataset	dataset	NOUN
ajst-16801	136	45	used	use	VERB
ajst-16801	136	46	.	.	PUNCT
ajst-16801	137	1	the	the	DET
ajst-16801	137	2	experimental	experimental	ADJ
ajst-16801	137	3	results	result	NOUN
ajst-16801	137	4	of	of	ADP
ajst-16801	137	5	dsdl	dsdl	NOUN
ajst-16801	137	6	in	in	ADP
ajst-16801	137	7	mushroom	mushroom	NOUN
ajst-16801	137	8	data	datum	NOUN
ajst-16801	137	9	set	set	VERB
ajst-16801	137	10	and	and	CCONJ
ajst-16801	137	11	other	other	ADJ
ajst-16801	137	12	10	10	NUM
ajst-16801	137	13	algorithms	algorithm	NOUN
ajst-16801	137	14	are	be	AUX
ajst-16801	137	15	listed	list	VERB
ajst-16801	137	16	in	in	ADP
ajst-16801	137	17	table	table	NOUN
ajst-16801	137	18	2	2	NUM
ajst-16801	137	19	,	,	PUNCT
ajst-16801	137	20	where	where	SCONJ
ajst-16801	137	21	the	the	DET
ajst-16801	137	22	best	good	ADJ
ajst-16801	137	23	results	result	NOUN
ajst-16801	137	24	are	be	AUX
ajst-16801	137	25	shown	show	VERB
ajst-16801	137	26	in	in	ADP
ajst-16801	137	27	bold	bold	ADJ
ajst-16801	137	28	.	.	PUNCT
ajst-16801	138	1	the	the	DET
ajst-16801	138	2	experimental	experimental	ADJ
ajst-16801	138	3	analysis	analysis	NOUN
ajst-16801	138	4	is	be	AUX
ajst-16801	138	5	as	as	SCONJ
ajst-16801	138	6	follows	follow	VERB
ajst-16801	138	7	:	:	PUNCT
ajst-16801	138	8	table	table	NOUN
ajst-16801	138	9	2	2	NUM
ajst-16801	138	10	.	.	PUNCT
ajst-16801	138	11	comparison	comparison	NOUN
ajst-16801	138	12	of	of	ADP
ajst-16801	138	13	classification	classification	NOUN
ajst-16801	138	14	effects	effect	NOUN
ajst-16801	138	15	of	of	ADP
ajst-16801	138	16	different	different	ADJ
ajst-16801	138	17	models	model	NOUN
ajst-16801	138	18	model	model	NOUN
ajst-16801	138	19	year	year	NOUN
ajst-16801	138	20	accuracy	accuracy	NOUN
ajst-16801	138	21	src	src	NOUN
ajst-16801	138	22	2009	2009	NUM
ajst-16801	138	23	86.72	86.72	NUM
ajst-16801	138	24	bsrc	bsrc	NOUN
ajst-16801	138	25	2011	2011	NUM
ajst-16801	138	26	91.93	91.93	NUM
ajst-16801	138	27	crc	crc	NOUN
ajst-16801	138	28	2011	2011	NUM
ajst-16801	138	29	93.80	93.80	NUM
ajst-16801	138	30	fddl	fddl	NOUN
ajst-16801	138	31	2011	2011	NUM
ajst-16801	138	32	96.81	96.81	NUM
ajst-16801	138	33	lc	lc	NOUN
ajst-16801	138	34	-	-	PUNCT
ajst-16801	138	35	ksvd	ksvd	NOUN
ajst-16801	138	36	2013	2013	NUM
ajst-16801	138	37	95.06	95.06	NUM
ajst-16801	138	38	svgdl	svgdl	NOUN
ajst-16801	138	39	2014	2014	NUM
ajst-16801	138	40	94.56	94.56	NUM
ajst-16801	138	41	dpl	dpl	NOUN
ajst-16801	138	42	2014	2014	NUM
ajst-16801	138	43	96.11	96.11	NUM
ajst-16801	138	44	erddpl	erddpl	NOUN
ajst-16801	138	45	2021	2021	NUM
ajst-16801	138	46	96.32	96.32	NUM
ajst-16801	138	47	slatdpl	slatdpl	NOUN
ajst-16801	138	48	2021	2021	NUM
ajst-16801	138	49	96.95	96.95	NUM
ajst-16801	138	50	svm	svm	ADJ
ajst-16801	138	51	-	-	ADJ
ajst-16801	138	52	ddpl	ddpl	ADJ
ajst-16801	138	53	2022	2022	NUM
ajst-16801	138	54	97.08	97.08	NUM
ajst-16801	138	55	dsdl	dsdl	PROPN
ajst-16801	138	56	2023	2023	NUM
ajst-16801	138	57	97.64	97.64	NUM
ajst-16801	138	58	as	as	SCONJ
ajst-16801	138	59	can	can	AUX
ajst-16801	138	60	be	be	AUX
ajst-16801	138	61	seen	see	VERB
ajst-16801	138	62	from	from	ADP
ajst-16801	138	63	table	table	NOUN
ajst-16801	138	64	2	2	NUM
ajst-16801	138	65	,	,	PUNCT
ajst-16801	138	66	the	the	DET
ajst-16801	138	67	classification	classification	NOUN
ajst-16801	138	68	accuracy	accuracy	NOUN
ajst-16801	138	69	of	of	ADP
ajst-16801	138	70	dsdl	dsdl	PROPN
ajst-16801	138	71	is	be	AUX
ajst-16801	138	72	0.56	0.56	NUM
ajst-16801	138	73	%	%	NOUN
ajst-16801	138	74	higher	high	ADJ
ajst-16801	138	75	than	than	ADP
ajst-16801	138	76	that	that	PRON
ajst-16801	138	77	of	of	ADP
ajst-16801	138	78	svm	svm	ADJ
ajst-16801	138	79	-	-	ADJ
ajst-16801	138	80	ddpl	ddpl	NOUN
ajst-16801	138	81	and	and	CCONJ
ajst-16801	138	82	0.69	0.69	NUM
ajst-16801	138	83	%	%	NOUN
ajst-16801	138	84	higher	high	ADJ
ajst-16801	138	85	than	than	ADP
ajst-16801	138	86	that	that	PRON
ajst-16801	138	87	of	of	ADP
ajst-16801	138	88	slatdpl	slatdpl	NOUN
ajst-16801	138	89	.	.	PUNCT
ajst-16801	139	1	these	these	DET
ajst-16801	139	2	data	datum	NOUN
ajst-16801	139	3	fully	fully	ADV
ajst-16801	139	4	demonstrate	demonstrate	VERB
ajst-16801	139	5	the	the	DET
ajst-16801	139	6	effective	effective	ADJ
ajst-16801	139	7	use	use	NOUN
ajst-16801	139	8	of	of	ADP
ajst-16801	139	9	stack	stack	NOUN
ajst-16801	139	10	autoencoder	autoencoder	NOUN
ajst-16801	139	11	in	in	ADP
ajst-16801	139	12	deep	deep	ADJ
ajst-16801	139	13	dictionary	dictionary	ADJ
ajst-16801	139	14	learning	learning	NOUN
ajst-16801	139	15	networks	network	NOUN
ajst-16801	139	16	.	.	PUNCT
ajst-16801	140	1	compared	compare	VERB
ajst-16801	140	2	with	with	ADP
ajst-16801	140	3	the	the	DET
ajst-16801	140	4	traditional	traditional	ADJ
ajst-16801	140	5	dictionary	dictionary	ADJ
ajst-16801	140	6	learning	learning	NOUN
ajst-16801	140	7	algorithm	algorithm	NOUN
ajst-16801	140	8	,	,	PUNCT
ajst-16801	140	9	the	the	DET
ajst-16801	140	10	classification	classification	NOUN
ajst-16801	140	11	accuracy	accuracy	NOUN
ajst-16801	140	12	of	of	ADP
ajst-16801	140	13	dsdl	dsdl	PROPN
ajst-16801	140	14	is	be	AUX
ajst-16801	140	15	improved	improve	VERB
ajst-16801	140	16	by	by	ADP
ajst-16801	140	17	0.81%~10.92	0.81%~10.92	PROPN
ajst-16801	140	18	%	%	NOUN
ajst-16801	140	19	,	,	PUNCT
ajst-16801	140	20	which	which	PRON
ajst-16801	140	21	further	far	ADV
ajst-16801	140	22	indicates	indicate	VERB
ajst-16801	140	23	that	that	SCONJ
ajst-16801	140	24	the	the	DET
ajst-16801	140	25	depth	depth	NOUN
ajst-16801	140	26	features	feature	NOUN
ajst-16801	140	27	obtained	obtain	VERB
ajst-16801	140	28	by	by	ADP
ajst-16801	140	29	the	the	DET
ajst-16801	140	30	algorithm	algorithm	NOUN
ajst-16801	140	31	are	be	AUX
ajst-16801	140	32	conducive	conducive	ADJ
ajst-16801	140	33	to	to	PART
ajst-16801	140	34	mushroom	mushroom	VERB
ajst-16801	140	35	image	image	NOUN
ajst-16801	140	36	classification	classification	NOUN
ajst-16801	140	37	.	.	PUNCT
ajst-16801	141	1	figure	figure	NOUN
ajst-16801	141	2	5	5	NUM
ajst-16801	141	3	.	.	PUNCT
ajst-16801	141	4	histogram	histogram	NOUN
ajst-16801	141	5	of	of	ADP
ajst-16801	141	6	classification	classification	NOUN
ajst-16801	141	7	results	result	NOUN
ajst-16801	141	8	of	of	ADP
ajst-16801	141	9	each	each	DET
ajst-16801	141	10	algorithm	algorithm	NOUN
ajst-16801	141	11	on	on	ADP
ajst-16801	141	12	mushroom	mushroom	NOUN
ajst-16801	141	13	data	datum	NOUN
ajst-16801	141	14	set	set	VERB
ajst-16801	141	15	it	it	PRON
ajst-16801	141	16	can	can	AUX
ajst-16801	141	17	be	be	AUX
ajst-16801	141	18	seen	see	VERB
ajst-16801	141	19	intuitively	intuitively	ADV
ajst-16801	141	20	from	from	ADP
ajst-16801	141	21	figure	figure	NOUN
ajst-16801	141	22	5	5	NUM
ajst-16801	141	23	that	that	SCONJ
ajst-16801	141	24	the	the	DET
ajst-16801	141	25	dsdl	dsdl	PROPN
ajst-16801	141	26	algorithm	algorithm	PROPN
ajst-16801	141	27	proposed	propose	VERB
ajst-16801	141	28	in	in	ADP
ajst-16801	141	29	this	this	DET
ajst-16801	141	30	paper	paper	NOUN
ajst-16801	141	31	has	have	VERB
ajst-16801	141	32	excellent	excellent	ADJ
ajst-16801	141	33	performance	performance	NOUN
ajst-16801	141	34	compared	compare	VERB
ajst-16801	141	35	with	with	ADP
ajst-16801	141	36	other	other	ADJ
ajst-16801	141	37	algorithms	algorithm	NOUN
ajst-16801	141	38	.	.	PUNCT
ajst-16801	142	1	in	in	ADP
ajst-16801	142	2	summary	summary	NOUN
ajst-16801	142	3	,	,	PUNCT
ajst-16801	142	4	from	from	ADP
ajst-16801	142	5	the	the	DET
ajst-16801	142	6	experimental	experimental	ADJ
ajst-16801	142	7	results	result	NOUN
ajst-16801	142	8	,	,	PUNCT
ajst-16801	142	9	it	it	PRON
ajst-16801	142	10	can	can	AUX
ajst-16801	142	11	be	be	AUX
ajst-16801	142	12	analyzed	analyze	VERB
ajst-16801	142	13	that	that	SCONJ
ajst-16801	142	14	:	:	PUNCT
ajst-16801	142	15	1	1	X
ajst-16801	142	16	.	.	X
ajst-16801	142	17	the	the	DET
ajst-16801	142	18	proposed	propose	VERB
ajst-16801	142	19	dsdl	dsdl	PROPN
ajst-16801	142	20	algorithm	algorithm	PROPN
ajst-16801	142	21	is	be	AUX
ajst-16801	142	22	slightly	slightly	ADV
ajst-16801	142	23	improved	improve	VERB
ajst-16801	142	24	than	than	ADP
ajst-16801	142	25	other	other	ADJ
ajst-16801	142	26	methods	method	NOUN
ajst-16801	142	27	and	and	CCONJ
ajst-16801	142	28	achieves	achieve	VERB
ajst-16801	142	29	higher	high	ADJ
ajst-16801	142	30	classification	classification	NOUN
ajst-16801	142	31	accuracy	accuracy	NOUN
ajst-16801	142	32	than	than	ADP
ajst-16801	142	33	traditional	traditional	ADJ
ajst-16801	142	34	dictionary	dictionary	NOUN
ajst-16801	142	35	-	-	PUNCT
ajst-16801	142	36	based	base	VERB
ajst-16801	142	37	image	image	NOUN
ajst-16801	142	38	classification	classification	NOUN
ajst-16801	142	39	methods	method	NOUN
ajst-16801	142	40	.	.	PUNCT
ajst-16801	143	1	traditional	traditional	ADJ
ajst-16801	143	2	dictionary	dictionary	ADJ
ajst-16801	143	3	learning	learning	NOUN
ajst-16801	143	4	methods	method	NOUN
ajst-16801	143	5	may	may	AUX
ajst-16801	143	6	not	not	PART
ajst-16801	143	7	be	be	AUX
ajst-16801	143	8	able	able	ADJ
ajst-16801	143	9	to	to	PART
ajst-16801	143	10	extract	extract	VERB
ajst-16801	143	11	robust	robust	ADJ
ajst-16801	143	12	representations	representation	NOUN
ajst-16801	143	13	.	.	PUNCT
ajst-16801	144	1	in	in	ADP
ajst-16801	144	2	contrast	contrast	NOUN
ajst-16801	144	3	,	,	PUNCT
ajst-16801	144	4	dsdl	dsdl	PROPN
ajst-16801	144	5	learns	learn	VERB
ajst-16801	144	6	both	both	CCONJ
ajst-16801	144	7	dictionary	dictionary	ADJ
ajst-16801	144	8	pairs	pair	NOUN
ajst-16801	144	9	and	and	CCONJ
ajst-16801	144	10	depth	depth	NOUN
ajst-16801	144	11	autoencoders	autoencoder	NOUN
ajst-16801	144	12	,	,	PUNCT
ajst-16801	144	13	which	which	PRON
ajst-16801	144	14	is	be	AUX
ajst-16801	144	15	more	more	ADV
ajst-16801	144	16	robust	robust	ADJ
ajst-16801	144	17	for	for	ADP
ajst-16801	144	18	image	image	NOUN
ajst-16801	144	19	classification	classification	NOUN
ajst-16801	144	20	.	.	PUNCT
ajst-16801	145	1	2	2	X
ajst-16801	145	2	.	.	X
ajst-16801	145	3	in	in	ADP
ajst-16801	145	4	general	general	ADJ
ajst-16801	145	5	,	,	PUNCT
ajst-16801	145	6	dsdl	dsdl	PROPN
ajst-16801	145	7	bridges	bridge	NOUN
ajst-16801	145	8	the	the	DET
ajst-16801	145	9	mutual	mutual	ADJ
ajst-16801	145	10	framework	framework	NOUN
ajst-16801	145	11	of	of	ADP
ajst-16801	145	12	traditional	traditional	ADJ
ajst-16801	145	13	discriminant	discriminant	ADJ
ajst-16801	145	14	dictionary	dictionary	ADJ
ajst-16801	145	15	learning	learning	NOUN
ajst-16801	145	16	and	and	CCONJ
ajst-16801	145	17	deep	deep	ADJ
ajst-16801	145	18	learning	learning	NOUN
ajst-16801	145	19	.	.	PUNCT
ajst-16801	146	1	the	the	DET
ajst-16801	146	2	traditional	traditional	ADJ
ajst-16801	146	3	dictionary	dictionary	ADJ
ajst-16801	146	4	learning	learning	NOUN
ajst-16801	146	5	method	method	NOUN
ajst-16801	146	6	presupposes	presuppose	VERB
ajst-16801	146	7	that	that	SCONJ
ajst-16801	146	8	images	image	NOUN
ajst-16801	146	9	are	be	AUX
ajst-16801	146	10	linear	linear	ADJ
ajst-16801	146	11	and	and	CCONJ
ajst-16801	146	12	can	can	AUX
ajst-16801	146	13	be	be	AUX
ajst-16801	146	14	represented	represent	VERB
ajst-16801	146	15	linearly	linearly	ADV
ajst-16801	146	16	by	by	ADP
ajst-16801	146	17	several	several	ADJ
ajst-16801	146	18	images	image	NOUN
ajst-16801	146	19	of	of	ADP
ajst-16801	146	20	the	the	DET
ajst-16801	146	21	same	same	ADJ
ajst-16801	146	22	class	class	NOUN
ajst-16801	146	23	,	,	PUNCT
ajst-16801	146	24	while	while	SCONJ
ajst-16801	146	25	dsdl	dsdl	NOUN
ajst-16801	146	26	makes	make	VERB
ajst-16801	146	27	images	image	NOUN
ajst-16801	146	28	more	more	ADV
ajst-16801	146	29	linearly	linearly	ADV
ajst-16801	146	30	correlated	correlate	VERB
ajst-16801	146	31	by	by	ADP
ajst-16801	146	32	converting	convert	VERB
ajst-16801	146	33	images	image	NOUN
ajst-16801	146	34	to	to	ADP
ajst-16801	146	35	deep	deep	ADJ
ajst-16801	146	36	space	space	NOUN
ajst-16801	146	37	.	.	PUNCT
ajst-16801	147	1	therefore	therefore	ADV
ajst-16801	147	2	,	,	PUNCT
ajst-16801	147	3	the	the	DET
ajst-16801	147	4	algorithm	algorithm	NOUN
ajst-16801	147	5	is	be	AUX
ajst-16801	147	6	more	more	ADV
ajst-16801	147	7	robust	robust	ADJ
ajst-16801	147	8	in	in	ADP
ajst-16801	147	9	image	image	NOUN
ajst-16801	147	10	classification	classification	NOUN
ajst-16801	147	11	based	base	VERB
ajst-16801	147	12	on	on	ADP
ajst-16801	147	13	dictionaries	dictionary	NOUN
ajst-16801	147	14	.	.	PUNCT
ajst-16801	148	1	4	4	X
ajst-16801	148	2	.	.	X
ajst-16801	148	3	conclusion	conclusion	NOUN
ajst-16801	148	4	in	in	ADP
ajst-16801	148	5	this	this	DET
ajst-16801	148	6	paper	paper	NOUN
ajst-16801	148	7	,	,	PUNCT
ajst-16801	148	8	we	we	PRON
ajst-16801	148	9	propose	propose	VERB
ajst-16801	148	10	a	a	DET
ajst-16801	148	11	small	small	ADJ
ajst-16801	148	12	sample	sample	NOUN
ajst-16801	148	13	mushroom	mushroom	NOUN
ajst-16801	148	14	image	image	NOUN
ajst-16801	148	15	classification	classification	NOUN
ajst-16801	148	16	recognition	recognition	NOUN
ajst-16801	148	17	algorithm	algorithm	NOUN
ajst-16801	148	18	,	,	PUNCT
ajst-16801	148	19	that	that	ADV
ajst-16801	148	20	is	is	ADV
ajst-16801	148	21	,	,	PUNCT
ajst-16801	148	22	deep	deep	ADV
ajst-16801	148	23	sparse	sparse	ADJ
ajst-16801	148	24	dictionary	dictionary	ADJ
ajst-16801	148	25	learning	learning	NOUN
ajst-16801	148	26	,	,	PUNCT
ajst-16801	148	27	which	which	PRON
ajst-16801	148	28	combines	combine	VERB
ajst-16801	148	29	dictionary	dictionary	ADJ
ajst-16801	148	30	learning	learn	VERB
ajst-16801	148	31	with	with	ADP
ajst-16801	148	32	autoencoder	autoencoder	NOUN
ajst-16801	148	33	framework	framework	NOUN
ajst-16801	148	34	.	.	PUNCT
ajst-16801	149	1	the	the	DET
ajst-16801	149	2	proposed	propose	VERB
ajst-16801	149	3	dsdl	dsdl	NOUN
ajst-16801	149	4	enables	enable	VERB
ajst-16801	149	5	us	we	PRON
ajst-16801	149	6	to	to	PART
ajst-16801	149	7	simultaneously	simultaneously	ADV
ajst-16801	149	8	learn	learn	VERB
ajst-16801	149	9	a	a	DET
ajst-16801	149	10	reconstruction	reconstruction	NOUN
ajst-16801	149	11	dictionary	dictionary	NOUN
ajst-16801	149	12	and	and	CCONJ
ajst-16801	149	13	an	an	DET
ajst-16801	149	14	analysis	analysis	NOUN
ajst-16801	149	15	dictionary	dictionary	NOUN
ajst-16801	149	16	of	of	ADP
ajst-16801	149	17	the	the	DET
ajst-16801	149	18	image	image	NOUN
ajst-16801	149	19	's	's	PART
ajst-16801	149	20	deep	deep	ADJ
ajst-16801	149	21	features	feature	NOUN
ajst-16801	149	22	,	,	PUNCT
ajst-16801	149	23	which	which	PRON
ajst-16801	149	24	contribute	contribute	VERB
ajst-16801	149	25	to	to	ADP
ajst-16801	149	26	a	a	DET
ajst-16801	149	27	robust	robust	ADJ
ajst-16801	149	28	representation	representation	NOUN
ajst-16801	149	29	.	.	PUNCT
ajst-16801	150	1	the	the	DET
ajst-16801	150	2	experimental	experimental	ADJ
ajst-16801	150	3	results	result	NOUN
ajst-16801	150	4	show	show	VERB
ajst-16801	150	5	that	that	SCONJ
ajst-16801	150	6	dsdl	dsdl	PROPN
ajst-16801	150	7	algorithm	algorithm	PROPN
ajst-16801	150	8	is	be	AUX
ajst-16801	150	9	better	well	ADJ
ajst-16801	150	10	than	than	ADP
ajst-16801	150	11	common	common	ADJ
ajst-16801	150	12	dictionary	dictionary	ADJ
ajst-16801	150	13	learning	learning	NOUN
ajst-16801	150	14	methods	method	NOUN
ajst-16801	150	15	in	in	ADP
ajst-16801	150	16	classifying	classify	VERB
ajst-16801	150	17	mushroom	mushroom	NOUN
ajst-16801	150	18	data	datum	NOUN
ajst-16801	150	19	set	set	VERB
ajst-16801	150	20	from	from	ADP
ajst-16801	150	21	kaggle	kaggle	PROPN
ajst-16801	150	22	.	.	PUNCT
ajst-16801	151	1	in	in	ADP
ajst-16801	151	2	terms	term	NOUN
ajst-16801	151	3	of	of	ADP
ajst-16801	151	4	application	application	NOUN
ajst-16801	151	5	,	,	PUNCT
ajst-16801	151	6	software	software	NOUN
ajst-16801	151	7	development	development	NOUN
ajst-16801	151	8	can	can	AUX
ajst-16801	151	9	be	be	AUX
ajst-16801	151	10	combined	combine	VERB
ajst-16801	151	11	with	with	ADP
ajst-16801	151	12	mobile	mobile	ADJ
ajst-16801	151	13	devices	device	NOUN
ajst-16801	151	14	and	and	CCONJ
ajst-16801	151	15	embedded	embed	VERB
ajst-16801	151	16	devices	device	NOUN
ajst-16801	151	17	in	in	ADP
ajst-16801	151	18	the	the	DET
ajst-16801	151	19	86	86	NUM
ajst-16801	151	20	87	87	NUM
ajst-16801	151	21	88	88	NUM
ajst-16801	151	22	89	89	NUM
ajst-16801	151	23	90	90	NUM
ajst-16801	151	24	91	91	NUM
ajst-16801	151	25	92	92	NUM
ajst-16801	151	26	93	93	NUM
ajst-16801	151	27	94	94	NUM
ajst-16801	151	28	95	95	NUM
ajst-16801	151	29	96	96	NUM
ajst-16801	151	30	97	97	NUM
ajst-16801	151	31	98	98	NUM
ajst-16801	151	32	cl	cl	NOUN
ajst-16801	151	33	as	as	ADP
ajst-16801	151	34	si	si	PROPN
ajst-16801	151	35	fic	fic	NOUN
ajst-16801	151	36	at	at	ADP
ajst-16801	151	37	io	io	PROPN
ajst-16801	151	38	n	n	ADV
ajst-16801	151	39	a	a	PRON
ajst-16801	151	40	cc	cc	X
ajst-16801	151	41	ur	ur	INTJ
ajst-16801	151	42	ac	ac	PROPN
ajst-16801	151	43	y	y	PROPN
ajst-16801	151	44	(	(	PUNCT
ajst-16801	151	45	%	%	NOUN
ajst-16801	151	46	)	)	PUNCT
ajst-16801	151	47	240	240	NUM
ajst-16801	151	48	future	future	NOUN
ajst-16801	151	49	,	,	PUNCT
ajst-16801	151	50	which	which	PRON
ajst-16801	151	51	will	will	AUX
ajst-16801	151	52	help	help	VERB
ajst-16801	151	53	improve	improve	VERB
ajst-16801	151	54	the	the	DET
ajst-16801	151	55	accuracy	accuracy	NOUN
ajst-16801	151	56	of	of	ADP
ajst-16801	151	57	classification	classification	NOUN
ajst-16801	151	58	of	of	ADP
ajst-16801	151	59	edible	edible	ADJ
ajst-16801	151	60	mushrooms	mushroom	NOUN
ajst-16801	151	61	and	and	CCONJ
ajst-16801	151	62	poisonous	poisonous	ADJ
ajst-16801	151	63	mushrooms	mushroom	NOUN
ajst-16801	151	64	,	,	PUNCT
ajst-16801	151	65	and	and	CCONJ
ajst-16801	151	66	further	far	ADV
ajst-16801	151	67	reduce	reduce	VERB
ajst-16801	151	68	the	the	DET
ajst-16801	151	69	occurrence	occurrence	NOUN
ajst-16801	151	70	of	of	ADP
ajst-16801	151	71	accidental	accidental	ADJ
ajst-16801	151	72	ingestion	ingestion	NOUN
ajst-16801	151	73	of	of	ADP
ajst-16801	151	74	poisonous	poisonous	ADJ
ajst-16801	151	75	mushrooms	mushroom	NOUN
ajst-16801	151	76	.	.	PUNCT
ajst-16801	152	1	references	reference	NOUN
ajst-16801	152	2	[	[	X
ajst-16801	152	3	1	1	NUM
ajst-16801	152	4	]	]	PUNCT
ajst-16801	152	5	chen	chen	PROPN
ajst-16801	152	6	y	y	PROPN
ajst-16801	152	7	,	,	PUNCT
ajst-16801	152	8	kalantidis	kalantidis	PROPN
ajst-16801	152	9	y	y	PROPN
ajst-16801	152	10	,	,	PUNCT
ajst-16801	152	11	li	li	PROPN
ajst-16801	152	12	j	j	PROPN
ajst-16801	152	13	,	,	PUNCT
ajst-16801	152	14	et	et	PROPN
ajst-16801	152	15	al	al	PROPN
ajst-16801	152	16	.	.	PUNCT
ajst-16801	153	1	multi	multi	ADJ
ajst-16801	153	2	-	-	ADJ
ajst-16801	153	3	fiber	fiber	ADJ
ajst-16801	153	4	networks	network	NOUN
ajst-16801	153	5	for	for	ADP
ajst-16801	153	6	video	video	NOUN
ajst-16801	153	7	recognition	recognition	NOUN
ajst-16801	154	1	[	[	X
ajst-16801	154	2	j	j	X
ajst-16801	154	3	]	]	X
ajst-16801	154	4	.	.	PUNCT
ajst-16801	155	1	2018	2018	NUM
ajst-16801	155	2	(	(	PUNCT
ajst-16801	155	3	1	1	NUM
ajst-16801	155	4	):	):	PUNCT
ajst-16801	155	5	364	364	NUM
ajst-16801	155	6	-	-	SYM
ajst-16801	155	7	380	380	NUM
ajst-16801	155	8	.	.	PUNCT
ajst-16801	156	1	[	[	X
ajst-16801	156	2	2	2	X
ajst-16801	156	3	]	]	PUNCT
ajst-16801	156	4	unnikrishnan	unnikrishnan	NOUN
ajst-16801	156	5	p	p	X
ajst-16801	156	6	,	,	PUNCT
ajst-16801	156	7	govindan	govindan	PROPN
ajst-16801	156	8	v	v	PROPN
ajst-16801	156	9	k	k	PROPN
ajst-16801	156	10	,	,	PUNCT
ajst-16801	156	11	kumar	kumar	PROPN
ajst-16801	156	12	s	s	PROPN
ajst-16801	156	13	d	d	PROPN
ajst-16801	156	14	m	m	PROPN
ajst-16801	156	15	.	.	PUNCT
ajst-16801	157	1	enhanced	enhance	VERB
ajst-16801	157	2	sparse	sparse	ADJ
ajst-16801	157	3	representation	representation	NOUN
ajst-16801	157	4	classifier	classifier	NOUN
ajst-16801	157	5	for	for	ADP
ajst-16801	157	6	text	text	NOUN
ajst-16801	157	7	classification[j	classification[j	NOUN
ajst-16801	157	8	]	]	X
ajst-16801	157	9	.	.	PUNCT
ajst-16801	158	1	expert	expert	NOUN
ajst-16801	158	2	systems	system	NOUN
ajst-16801	158	3	with	with	ADP
ajst-16801	158	4	application	application	NOUN
ajst-16801	158	5	,	,	PUNCT
ajst-16801	158	6	2019	2019	NUM
ajst-16801	158	7	,	,	PUNCT
ajst-16801	158	8	129(sep.):260	129(sep.):260	NUM
ajst-16801	158	9	-	-	SYM
ajst-16801	158	10	272	272	NUM
ajst-16801	158	11	.	.	PUNCT
ajst-16801	159	1	[	[	X
ajst-16801	159	2	3	3	X
ajst-16801	159	3	]	]	PUNCT
ajst-16801	159	4	l.shen	l.shen	ADV
ajst-16801	159	5	,	,	PUNCT
ajst-16801	159	6	g.sun	g.sun	PROPN
ajst-16801	159	7	,	,	PUNCT
ajst-16801	159	8	q.huang	q.huang	X
ajst-16801	159	9	,	,	PUNCT
ajst-16801	159	10	s.wang	s.wang	X
ajst-16801	159	11	,	,	PUNCT
ajst-16801	159	12	z.lin	z.lin	NOUN
ajst-16801	159	13	,	,	PUNCT
ajst-16801	159	14	e.wu.multi	e.wu.multi	ADJ
ajst-16801	159	15	-	-	PUNCT
ajst-16801	159	16	level	level	NOUN
ajst-16801	159	17	discriminative	discriminative	NOUN
ajst-16801	159	18	dictionary	dictionary	NOUN
ajst-16801	159	19	learning	learn	VERB
ajst-16801	159	20	with	with	ADP
ajst-16801	159	21	application	application	NOUN
ajst-16801	159	22	to	to	ADP
ajst-16801	159	23	large	large	ADJ
ajst-16801	159	24	scale	scale	NOUN
ajst-16801	159	25	image	image	NOUN
ajst-16801	159	26	classification[j	classification[j	NOUN
ajst-16801	159	27	]	]	PUNCT
ajst-16801	159	28	.	.	PUNCT
ajst-16801	160	1	ieee	ieee	NOUN
ajst-16801	160	2	transactions	transaction	NOUN
ajst-16801	160	3	on	on	ADP
ajst-16801	160	4	image	image	NOUN
ajst-16801	160	5	processing	processing	NOUN
ajst-16801	160	6	,	,	PUNCT
ajst-16801	160	7	2015	2015	NUM
ajst-16801	160	8	,	,	PUNCT
ajst-16801	160	9	24(10	24(10	NUM
ajst-16801	160	10	):	):	PUNCT
ajst-16801	160	11	3109	3109	NUM
ajst-16801	160	12	-	-	SYM
ajst-16801	160	13	3123	3123	NUM
ajst-16801	160	14	.	.	PUNCT
ajst-16801	161	1	[	[	X
ajst-16801	161	2	4	4	NUM
ajst-16801	161	3	]	]	X
ajst-16801	161	4	s.bahrampour	s.bahrampour	ADJ
ajst-16801	161	5	,	,	PUNCT
ajst-16801	161	6	n.m.nasrabadi	n.m.nasrabadi	NOUN
ajst-16801	161	7	,	,	PUNCT
ajst-16801	161	8	a.ray	a.ray	NOUN
ajst-16801	161	9	,	,	PUNCT
ajst-16801	161	10	w.k.jenkins	w.k.jenkin	NOUN
ajst-16801	161	11	.	.	PUNCT
ajst-16801	162	1	multimodal	multimodal	NOUN
ajst-16801	162	2	task	task	NOUN
ajst-16801	162	3	-	-	PUNCT
ajst-16801	162	4	driven	drive	VERB
ajst-16801	162	5	dictionary	dictionary	NOUN
ajst-16801	162	6	learning	learning	NOUN
ajst-16801	162	7	for	for	ADP
ajst-16801	162	8	image	image	NOUN
ajst-16801	162	9	classification[j	classification[j	NOUN
ajst-16801	162	10	]	]	PUNCT
ajst-16801	162	11	.	.	PUNCT
ajst-16801	163	1	ieee	ieee	NOUN
ajst-16801	163	2	transactions	transaction	NOUN
ajst-16801	163	3	on	on	ADP
ajst-16801	163	4	image	image	NOUN
ajst-16801	163	5	processing,2015	processing,2015	PROPN
ajst-16801	163	6	,	,	PUNCT
ajst-16801	163	7	25(1):24	25(1):24	NUM
ajst-16801	163	8	-	-	SYM
ajst-16801	163	9	38	38	NUM
ajst-16801	163	10	.	.	PUNCT
ajst-16801	164	1	[	[	X
ajst-16801	164	2	5	5	NUM
ajst-16801	164	3	]	]	X
ajst-16801	164	4	mairal	mairal	ADJ
ajst-16801	164	5	,	,	PUNCT
ajst-16801	164	6	f.	f.	PROPN
ajst-16801	164	7	bach	bach	PROPN
ajst-16801	164	8	,	,	PUNCT
ajst-16801	164	9	j.	j.	PROPN
ajst-16801	164	10	poncd	poncd	PROPN
ajst-16801	164	11	,	,	PUNCT
ajst-16801	164	12	g.	g.	PROPN
ajst-16801	164	13	sapior	sapior	PROPN
ajst-16801	164	14	.	.	PUNCT
ajst-16801	165	1	online	online	PROPN
ajst-16801	165	2	dictionary	dictionary	ADJ
ajst-16801	165	3	learning	learn	VERB
ajst-16801	165	4	for	for	ADP
ajst-16801	165	5	sparse	sparse	ADJ
ajst-16801	165	6	coding[c	coding[c	NOUN
ajst-16801	165	7	]	]	PUNCT
ajst-16801	165	8	.	.	PUNCT
ajst-16801	166	1	proceedings	proceeding	NOUN
ajst-16801	166	2	of	of	ADP
ajst-16801	166	3	the	the	DET
ajst-16801	166	4	26th	26th	ADJ
ajst-16801	166	5	annual	annual	ADJ
ajst-16801	166	6	international	international	ADJ
ajst-16801	166	7	conference	conference	NOUN
ajst-16801	166	8	on	on	ADP
ajst-16801	166	9	machine	machine	NOUN
ajst-16801	166	10	learninb	learninb	NOUN
ajst-16801	166	11	.	.	PUNCT
ajst-16801	167	1	2009:689	2009:689	NUM
ajst-16801	167	2	-	-	PUNCT
ajst-16801	167	3	696	696	NUM
ajst-16801	167	4	.	.	PUNCT
ajst-16801	168	1	[	[	X
ajst-16801	168	2	6	6	NUM
ajst-16801	168	3	]	]	PUNCT
ajst-16801	168	4	h.zhang	h.zhang	NOUN
ajst-16801	168	5	,	,	PUNCT
ajst-16801	168	6	liu	liu	PROPN
ajst-16801	168	7	h	h	PROPN
ajst-16801	168	8	,	,	PUNCT
ajst-16801	168	9	song	song	NOUN
ajst-16801	168	10	and	and	CCONJ
ajst-16801	168	11	r.f.sun	r.f.sun	PROPN
ajst-16801	168	12	.	.	PUNCT
ajst-16801	169	1	nonlinear	nonlinear	ADJ
ajst-16801	169	2	dictionary	dictionary	PROPN
ajst-16801	169	3	learning	learning	PROPN
ajst-16801	169	4	based	base	VERB
ajst-16801	169	5	deep	deep	ADJ
ajst-16801	169	6	neural	neural	NOUN
ajst-16801	169	7	networks[c	networks[c	PROPN
ajst-16801	169	8	]	]	PUNCT
ajst-16801	169	9	.	.	PUNCT
ajst-16801	170	1	2016	2016	NUM
ajst-16801	170	2	international	international	ADJ
ajst-16801	170	3	joint	joint	ADJ
ajst-16801	170	4	conference	conference	NOUN
ajst-16801	170	5	on	on	ADP
ajst-16801	170	6	neural	neural	ADJ
ajst-16801	170	7	networks	network	NOUN
ajst-16801	170	8	(	(	PUNCT
ajst-16801	170	9	ijcnn	ijcnn	PROPN
ajst-16801	170	10	)	)	PUNCT
ajst-16801	170	11	.	.	PUNCT
ajst-16801	171	1	ieee,2016	ieee,2016	NOUN
ajst-16801	171	2	:	:	PUNCT
ajst-16801	171	3	3771	3771	NUM
ajst-16801	171	4	-	-	SYM
ajst-16801	171	5	3776	3776	NUM
ajst-16801	171	6	.	.	PUNCT
ajst-16801	172	1	[	[	X
ajst-16801	172	2	7	7	X
ajst-16801	172	3	]	]	X
ajst-16801	172	4	s.tariyal	s.tariyal	NOUN
ajst-16801	172	5	,	,	PUNCT
ajst-16801	172	6	h.	h.	NOUN
ajst-16801	172	7	aggarwal	aggarwal	PROPN
ajst-16801	172	8	,	,	PUNCT
ajst-16801	172	9	a.majumdar	a.majumdar	ADV
ajst-16801	172	10	.	.	PUNCT
ajst-16801	173	1	greedy	greedy	ADJ
ajst-16801	173	2	deep	deep	ADJ
ajst-16801	173	3	dictionary	dictionary	ADJ
ajst-16801	173	4	learning	learn	VERB
ajst-16801	173	5	for	for	ADP
ajst-16801	173	6	hyperspectral	hyperspectral	ADJ
ajst-16801	173	7	image	image	NOUN
ajst-16801	173	8	classification	classification	NOUN
ajst-16801	173	9	[	[	X
ajst-16801	173	10	c].2016	c].2016	PROPN
ajst-16801	173	11	8th	8th	ADJ
ajst-16801	173	12	workshop	workshop	NOUN
ajst-16801	173	13	on	on	ADP
ajst-16801	173	14	hyperspectral	hyperspectral	ADJ
ajst-16801	173	15	image	image	NOUN
ajst-16801	173	16	and	and	CCONJ
ajst-16801	173	17	signal	signal	NOUN
ajst-16801	173	18	processing	processing	NOUN
ajst-16801	173	19	:	:	PUNCT
ajst-16801	173	20	evolution	evolution	NOUN
ajst-16801	173	21	in	in	ADP
ajst-16801	173	22	remote	remote	ADJ
ajst-16801	173	23	sensing	sensing	NOUN
ajst-16801	173	24	(	(	PUNCT
ajst-16801	173	25	whispers	whisper	NOUN
ajst-16801	173	26	)	)	PUNCT
ajst-16801	173	27	.	.	PUNCT
ajst-16801	174	1	ieee	ieee	PROPN
ajst-16801	174	2	,	,	PUNCT
ajst-16801	174	3	2016	2016	NUM
ajst-16801	174	4	:	:	PUNCT
ajst-16801	174	5	1	1	NUM
ajst-16801	174	6	-	-	SYM
ajst-16801	174	7	4	4	NUM
ajst-16801	174	8	.	.	PUNCT
ajst-16801	175	1	[	[	X
ajst-16801	175	2	8	8	NUM
ajst-16801	175	3	]	]	X
ajst-16801	175	4	v.papyan	v.papyan	ADJ
ajst-16801	175	5	,	,	PUNCT
ajst-16801	175	6	y.romano	y.romano	PRON
ajst-16801	175	7	,	,	PUNCT
ajst-16801	175	8	j.sulam	j.sulam	PROPN
ajst-16801	175	9	,	,	PUNCT
ajst-16801	175	10	m.elad	m.elad	NOUN
ajst-16801	175	11	.	.	PUNCT
ajst-16801	176	1	convolutional	convolutional	ADJ
ajst-16801	176	2	　	　	SPACE
ajst-16801	176	3	dictionary	dictionary	ADJ
ajst-16801	176	4	learning	learning	NOUN
ajst-16801	176	5	via	via	ADP
ajst-16801	176	6	local	local	ADJ
ajst-16801	176	7	processing[c	processing[c	NOUN
ajst-16801	176	8	]	]	PUNCT
ajst-16801	176	9	.	.	PUNCT
ajst-16801	177	1	proceedings	proceeding	NOUN
ajst-16801	177	2	of	of	ADP
ajst-16801	177	3	the	the	DET
ajst-16801	177	4	ieee	ieee	NOUN
ajst-16801	177	5	international	international	PROPN
ajst-16801	177	6	conference	conference	NOUN
ajst-16801	177	7	on	on	ADP
ajst-16801	177	8	　	　	SPACE
ajst-16801	177	9	computer	computer	NOUN
ajst-16801	177	10	vision	vision	NOUN
ajst-16801	177	11	.	.	PUNCT
ajst-16801	178	1	2017	2017	NUM
ajst-16801	178	2	:	:	PUNCT
ajst-16801	178	3	5296	5296	NUM
ajst-16801	178	4	-	-	SYM
ajst-16801	178	5	5304	5304	NUM
ajst-16801	178	6	.	.	PUNCT
ajst-16801	179	1	[	[	X
ajst-16801	179	2	9	9	NUM
ajst-16801	179	3	]	]	SYM
ajst-16801	179	4	j.yang	j.yang	PROPN
ajst-16801	179	5	,	,	PUNCT
ajst-16801	179	6	m.h.yang	m.h.yang	PROPN
ajst-16801	179	7	.	.	PUNCT
ajst-16801	179	8	top	top	ADJ
ajst-16801	179	9	-	-	PUNCT
ajst-16801	179	10	down	down	ADP
ajst-16801	179	11	visual	visual	ADJ
ajst-16801	179	12	saliency	saliency	NOUN
ajst-16801	179	13	via	via	ADP
ajst-16801	179	14	joint	joint	ADJ
ajst-16801	179	15	　	　	SPACE
ajst-16801	179	16	crf	crf	PROPN
ajst-16801	179	17	and	and	CCONJ
ajst-16801	179	18	dictionary	dictionary	PROPN
ajst-16801	179	19	leaming[j	leaming[j	PROPN
ajst-16801	179	20	]	]	PUNCT
ajst-16801	179	21	.	.	PUNCT
ajst-16801	180	1	ieee	ieee	NOUN
ajst-16801	180	2	transactions	transaction	NOUN
ajst-16801	180	3	on	on	ADP
ajst-16801	180	4	　	　	SPACE
ajst-16801	180	5	pattern	pattern	NOUN
ajst-16801	180	6	analysis	analysis	NOUN
ajst-16801	180	7	and	and	CCONJ
ajst-16801	180	8	machine	machine	NOUN
ajst-16801	180	9	intelligence	intelligence	NOUN
ajst-16801	180	10	,	,	PUNCT
ajst-16801	180	11	2016	2016	NUM
ajst-16801	180	12	,	,	PUNCT
ajst-16801	180	13	39(3	39(3	NUM
ajst-16801	180	14	):	):	PUNCT
ajst-16801	180	15	　	　	SPACE
ajst-16801	180	16	576	576	NUM
ajst-16801	180	17	-	-	SYM
ajst-16801	180	18	588	588	NUM
ajst-16801	180	19	.	.	PUNCT
ajst-16801	181	1	[	[	X
ajst-16801	181	2	10	10	NUM
ajst-16801	181	3	]	]	PUNCT
ajst-16801	181	4	v.singhal	v.singhal	NOUN
ajst-16801	181	5	,	,	PUNCT
ajst-16801	181	6	a.majumdar	a.majumdar	NOUN
ajst-16801	181	7	.	.	PUNCT
ajst-16801	182	1	majorization	majorization	NOUN
ajst-16801	182	2	minimization	minimization	NOUN
ajst-16801	182	3	technique	technique	NOUN
ajst-16801	182	4	for	for	ADP
ajst-16801	182	5	optimally	optimally	ADV
ajst-16801	182	6	solving	solve	VERB
ajst-16801	182	7	deep	deep	ADJ
ajst-16801	182	8	dictionary	dictionary	ADJ
ajst-16801	182	9	learning[j	learning[j	NOUN
ajst-16801	182	10	]	]	PUNCT
ajst-16801	182	11	.	.	PUNCT
ajst-16801	183	1	neural	neural	ADJ
ajst-16801	183	2	processing	processing	NOUN
ajst-16801	183	3	letters	letter	NOUN
ajst-16801	183	4	,	,	PUNCT
ajst-16801	183	5	2018	2018	NUM
ajst-16801	183	6	,	,	PUNCT
ajst-16801	183	7	47(3	47(3	NUM
ajst-16801	183	8	):	):	PUNCT
ajst-16801	183	9	799	799	NUM
ajst-16801	183	10	-	-	SYM
ajst-16801	183	11	814	814	NUM
ajst-16801	183	12	.	.	PUNCT
ajst-16801	184	1	[	[	X
ajst-16801	184	2	11	11	NUM
ajst-16801	184	3	]	]	SYM
ajst-16801	184	4	y.liu	y.liu	PROPN
ajst-16801	184	5	,	,	PUNCT
ajst-16801	184	6	q.chen	q.chen	ADV
ajst-16801	184	7	,	,	PUNCT
ajst-16801	184	8	w.chen	w.chen	ADV
ajst-16801	184	9	,	,	PUNCT
ajst-16801	184	10	i.	i.	PROPN
ajst-16801	184	11	wassell	wassell	PROPN
ajst-16801	184	12	.	.	PUNCT
ajst-16801	185	1	dictionary	dictionary	ADJ
ajst-16801	185	2	learning	learning	NOUN
ajst-16801	185	3	inspired	inspire	VERB
ajst-16801	185	4	deep	deep	ADJ
ajst-16801	185	5	network	network	NOUN
ajst-16801	185	6	for	for	ADP
ajst-16801	185	7	scene	scene	NOUN
ajst-16801	185	8	recognition[c	recognition[c	PROPN
ajst-16801	185	9	]	]	PUNCT
ajst-16801	185	10	.	.	PUNCT
ajst-16801	186	1	thirty	thirty	NUM
ajst-16801	186	2	-	-	PUNCT
ajst-16801	186	3	second	second	NOUN
ajst-16801	186	4	aaai	aaai	PROPN
ajst-16801	186	5	conference	conference	NOUN
ajst-16801	186	6	on	on	ADP
ajst-16801	186	7	artificial	artificial	ADJ
ajst-16801	186	8	intelligence	intelligence	NOUN
ajst-16801	186	9	.	.	PUNCT
ajst-16801	187	1	2018	2018	NUM
ajst-16801	187	2	.	.	PUNCT
ajst-16801	188	1	[	[	X
ajst-16801	188	2	12	12	NUM
ajst-16801	188	3	]	]	PUNCT
ajst-16801	188	4	j.sulam	j.sulam	PROPN
ajst-16801	188	5	,	,	PUNCT
ajst-16801	188	6	v.papyan	v.papyan	ADJ
ajst-16801	188	7	,	,	PUNCT
ajst-16801	188	8	y.romano	y.romano	NOUN
ajst-16801	188	9	,	,	PUNCT
ajst-16801	188	10	m.elad	m.elad	NOUN
ajst-16801	188	11	.	.	PUNCT
ajst-16801	189	1	multilayer	multilayer	ADJ
ajst-16801	189	2	convolutional	convolutional	ADJ
ajst-16801	189	3	sparse	sparse	ADJ
ajst-16801	189	4	modeling	modeling	NOUN
ajst-16801	189	5	and	and	CCONJ
ajst-16801	189	6	dictionary	dictionary	ADJ
ajst-16801	189	7	learning[j	learning[j	PROPN
ajst-16801	189	8	]	]	PUNCT
ajst-16801	189	9	.	.	PUNCT
ajst-16801	190	1	ieee	ieee	NOUN
ajst-16801	190	2	transactions	transaction	NOUN
ajst-16801	190	3	on	on	ADP
ajst-16801	190	4	signal	signal	ADJ
ajst-16801	190	5	processing	processing	NOUN
ajst-16801	190	6	,	,	PUNCT
ajst-16801	190	7	2018,66(15	2018,66(15	NUM
ajst-16801	190	8	):	):	PUNCT
ajst-16801	190	9	40904104	40904104	NUM
ajst-16801	190	10	.	.	PUNCT
ajst-16801	191	1	[	[	X
ajst-16801	191	2	13	13	NUM
ajst-16801	191	3	]	]	PUNCT
ajst-16801	191	4	j.	j.	PROPN
ajst-16801	191	5	hu	hu	PROPN
ajst-16801	191	6	,	,	PUNCT
ajst-16801	191	7	y.p	y.p	PROPN
ajst-16801	191	8	.	.	PROPN
ajst-16801	191	9	tan	tan	PROPN
ajst-16801	191	10	.	.	PUNCT
ajst-16801	192	1	nonlinear	nonlinear	ADJ
ajst-16801	192	2	dictionary	dictionary	PROPN
ajst-16801	192	3	learning	learning	NOUN
ajst-16801	192	4	with	with	ADP
ajst-16801	192	5	application	application	NOUN
ajst-16801	192	6	to	to	ADP
ajst-16801	192	7	image	image	NOUN
ajst-16801	192	8	classification[j	classification[j	NOUN
ajst-16801	192	9	]	]	PUNCT
ajst-16801	192	10	.	.	PUNCT
ajst-16801	193	1	pattern	pattern	NOUN
ajst-16801	193	2	recognition	recognition	NOUN
ajst-16801	193	3	,	,	PUNCT
ajst-16801	193	4	2018	2018	NUM
ajst-16801	193	5	,	,	PUNCT
ajst-16801	193	6	75:282291	75:282291	NUM
ajst-16801	193	7	.	.	PUNCT
ajst-16801	194	1	[	[	X
ajst-16801	194	2	14	14	NUM
ajst-16801	194	3	]	]	SYM
ajst-16801	194	4	h.tang	h.tang	PROPN
ajst-16801	194	5	,	,	PUNCT
ajst-16801	194	6	h.wei	h.wei	PROPN
ajst-16801	194	7	,	,	PUNCT
ajst-16801	194	8	w.xiao.deep	w.xiao.deep	PROPN
ajst-16801	194	9	micro	micro	NOUN
ajst-16801	194	10	-	-	NOUN
ajst-16801	194	11	dictionary	dictionary	ADJ
ajst-16801	194	12	learning	learning	NOUN
ajst-16801	194	13	and	and	CCONJ
ajst-16801	194	14	coding	code	VERB
ajst-16801	194	15	network[c].2019	network[c].2019	ADP
ajst-16801	194	16	ieee	ieee	PROPN
ajst-16801	194	17	winter	winter	NOUN
ajst-16801	194	18	conference	conference	NOUN
ajst-16801	194	19	on	on	ADP
ajst-16801	194	20	applications	application	NOUN
ajst-16801	194	21	of	of	ADP
ajst-16801	194	22	computer	computer	NOUN
ajst-16801	194	23	vision	vision	NOUN
ajst-16801	194	24	(	(	PUNCT
ajst-16801	194	25	wacv	wacv	NOUN
ajst-16801	194	26	)	)	PUNCT
ajst-16801	194	27	.	.	PUNCT
ajst-16801	195	1	ieee	ieee	NOUN
ajst-16801	195	2	,	,	PUNCT
ajst-16801	195	3	2019	2019	NUM
ajst-16801	195	4	:	:	PUNCT
ajst-16801	195	5	386395	386395	NUM
ajst-16801	195	6	.	.	PUNCT
ajst-16801	196	1	[	[	X
ajst-16801	196	2	15	15	NUM
ajst-16801	196	3	]	]	PUNCT
ajst-16801	196	4	engan	engan	NOUN
ajst-16801	196	5	k	k	PROPN
ajst-16801	196	6	,	,	PUNCT
ajst-16801	196	7	aase	aase	PROPN
ajst-16801	196	8	s	s	PART
ajst-16801	196	9	o	o	PROPN
ajst-16801	196	10	,	,	PUNCT
ajst-16801	196	11	husoy	husoy	PROPN
ajst-16801	196	12	j	j	PROPN
ajst-16801	196	13	h.	h.	PROPN
ajst-16801	196	14	frame	frame	PROPN
ajst-16801	196	15	based	base	VERB
ajst-16801	196	16	signal	signal	NOUN
ajst-16801	196	17	compression	compression	NOUN
ajst-16801	196	18	using	use	VERB
ajst-16801	196	19	method	method	NOUN
ajst-16801	196	20	of	of	ADP
ajst-16801	196	21	optimal	optimal	ADJ
ajst-16801	196	22	directions	direction	NOUN
ajst-16801	196	23	(	(	PUNCT
ajst-16801	196	24	mod)[c]//iscas'99	mod)[c]//iscas'99	NOUN
ajst-16801	196	25	.	.	PUNCT
ajst-16801	196	26	proceedings	proceeding	NOUN
ajst-16801	196	27	of	of	ADP
ajst-16801	196	28	the	the	DET
ajst-16801	196	29	1999	1999	NUM
ajst-16801	196	30	ieee	ieee	NOUN
ajst-16801	196	31	international	international	ADJ
ajst-16801	196	32	symposium	symposium	NOUN
ajst-16801	196	33	on	on	ADP
ajst-16801	196	34	circuits	circuit	NOUN
ajst-16801	196	35	and	and	CCONJ
ajst-16801	196	36	systems	system	NOUN
ajst-16801	196	37	vlsi	vlsi	PROPN
ajst-16801	196	38	(	(	PUNCT
ajst-16801	196	39	cat	cat	NOUN
ajst-16801	196	40	.	.	PUNCT
ajst-16801	197	1	no	no	INTJ
ajst-16801	197	2	.	.	NOUN
ajst-16801	197	3	99ch36349	99ch36349	NUM
ajst-16801	197	4	)	)	PUNCT
ajst-16801	197	5	.	.	PUNCT
ajst-16801	198	1	ieee	ieee	PROPN
ajst-16801	198	2	,	,	PUNCT
ajst-16801	198	3	1999	1999	NUM
ajst-16801	198	4	,	,	PUNCT
ajst-16801	198	5	4	4	NUM
ajst-16801	198	6	:	:	SYM
ajst-16801	198	7	1	1	NUM
ajst-16801	198	8	-	-	SYM
ajst-16801	198	9	4	4	NUM
ajst-16801	198	10	.	.	PUNCT
ajst-16801	199	1	[	[	X
ajst-16801	199	2	16	16	NUM
ajst-16801	199	3	]	]	X
ajst-16801	199	4	mairal	mairal	PROPN
ajst-16801	199	5	j	j	PROPN
ajst-16801	199	6	,	,	PUNCT
ajst-16801	199	7	bach	bach	PROPN
ajst-16801	199	8	f	f	PROPN
ajst-16801	199	9	,	,	PUNCT
ajst-16801	199	10	ponce	ponce	PROPN
ajst-16801	199	11	j	j	PROPN
ajst-16801	199	12	,	,	PUNCT
ajst-16801	199	13	et	et	PROPN
ajst-16801	199	14	al	al	PROPN
ajst-16801	199	15	.	.	PROPN
ajst-16801	200	1	online	online	PROPN
ajst-16801	200	2	dictionary	dictionary	PROPN
ajst-16801	200	3	learning	learn	VERB
ajst-16801	200	4	for	for	ADP
ajst-16801	200	5	sparse	sparse	ADJ
ajst-16801	200	6	coding[c]//proceedings	coding[c]//proceeding	NOUN
ajst-16801	200	7	of	of	ADP
ajst-16801	200	8	the	the	DET
ajst-16801	200	9	26th	26th	ADJ
ajst-16801	200	10	annual	annual	ADJ
ajst-16801	200	11	international	international	ADJ
ajst-16801	200	12	conference	conference	NOUN
ajst-16801	200	13	on	on	ADP
ajst-16801	200	14	machine	machine	NOUN
ajst-16801	200	15	learning	learning	NOUN
ajst-16801	200	16	.	.	PUNCT
ajst-16801	201	1	acm	acm	PROPN
ajst-16801	201	2	,	,	PUNCT
ajst-16801	201	3	2009	2009	NUM
ajst-16801	201	4	:	:	PUNCT
ajst-16801	201	5	689	689	NUM
ajst-16801	201	6	-	-	SYM
ajst-16801	201	7	696	696	NUM
ajst-16801	201	8	.	.	PUNCT
ajst-16801	202	1	[	[	X
ajst-16801	202	2	17	17	NUM
ajst-16801	202	3	]	]	X
ajst-16801	202	4	aharon	aharon	PROPN
ajst-16801	202	5	m	m	PROPN
ajst-16801	202	6	,	,	PUNCT
ajst-16801	202	7	elad	elad	PROPN
ajst-16801	202	8	m	m	PROPN
ajst-16801	202	9	,	,	PUNCT
ajst-16801	202	10	bruckstein	bruckstein	ADJ
ajst-16801	202	11	a.	a.	NOUN
ajst-16801	202	12	k	k	PROPN
ajst-16801	202	13	-	-	PUNCT
ajst-16801	202	14	svd	svd	PROPN
ajst-16801	202	15	:	:	PUNCT
ajst-16801	202	16	an	an	DET
ajst-16801	202	17	algorithm	algorithm	NOUN
ajst-16801	202	18	for	for	ADP
ajst-16801	202	19	designing	design	VERB
ajst-16801	202	20	overcomplete	overcomplete	ADJ
ajst-16801	202	21	dictionaries	dictionary	NOUN
ajst-16801	202	22	for	for	ADP
ajst-16801	202	23	sparse	sparse	ADJ
ajst-16801	202	24	representation[j	representation[j	PROPN
ajst-16801	202	25	]	]	PUNCT
ajst-16801	202	26	.	.	PUNCT
ajst-16801	203	1	ieee	ieee	NOUN
ajst-16801	203	2	transactions	transaction	NOUN
ajst-16801	203	3	on	on	ADP
ajst-16801	203	4	signal	signal	ADJ
ajst-16801	203	5	processing	processing	NOUN
ajst-16801	203	6	,	,	PUNCT
ajst-16801	203	7	2006	2006	NUM
ajst-16801	203	8	,	,	PUNCT
ajst-16801	203	9	54(11):4311	54(11):4311	NUM
ajst-16801	203	10	-	-	SYM
ajst-16801	203	11	4322	4322	NUM
ajst-16801	203	12	.	.	PUNCT
ajst-16801	204	1	[	[	X
ajst-16801	204	2	18	18	NUM
ajst-16801	204	3	]	]	PUNCT
ajst-16801	204	4	krizhevsky	krizhevsky	NOUN
ajst-16801	204	5	a	a	PROPN
ajst-16801	204	6	,	,	PUNCT
ajst-16801	204	7	sutskeveri	sutskeveri	PROPN
ajst-16801	204	8	,	,	PUNCT
ajst-16801	204	9	hinton	hinton	PROPN
ajst-16801	204	10	g(2012)imagenet	g(2012)imagenet	NOUN
ajst-16801	204	11	classification	classification	NOUN
ajst-16801	204	12	with	with	ADP
ajst-16801	204	13	deep	deep	ADJ
ajst-16801	204	14	convolutional	convolutional	ADJ
ajst-16801	204	15	neural	neural	ADJ
ajst-16801	204	16	networks	network	NOUN
ajst-16801	204	17	.	.	PUNCT
ajst-16801	205	1	[	[	X
ajst-16801	205	2	19	19	NUM
ajst-16801	205	3	]	]	PUNCT
ajst-16801	205	4	wright	wright	PROPN
ajst-16801	205	5	j	j	PROPN
ajst-16801	205	6	,	,	PUNCT
ajst-16801	205	7	yang	yang	PROPN
ajst-16801	205	8	ay	ay	PROPN
ajst-16801	205	9	,	,	PUNCT
ajst-16801	205	10	ganesh	ganesh	PROPN
ajst-16801	205	11	a	a	PROPN
ajst-16801	205	12	,	,	PUNCT
ajst-16801	205	13	sastry	sastry	PROPN
ajst-16801	205	14	ss	ss	PROPN
ajst-16801	205	15	,	,	PUNCT
ajst-16801	205	16	ma	ma	PROPN
ajst-16801	205	17	y	y	PROPN
ajst-16801	205	18	(	(	PUNCT
ajst-16801	205	19	2009	2009	NUM
ajst-16801	205	20	)	)	PUNCT
ajst-16801	205	21	robust	robust	ADJ
ajst-16801	205	22	face	face	NOUN
ajst-16801	205	23	recognition	recognition	NOUN
ajst-16801	205	24	via	via	ADP
ajst-16801	205	25	sparse	sparse	ADJ
ajst-16801	205	26	representation	representation	NOUN
ajst-16801	205	27	.	.	PUNCT
ajst-16801	206	1	ieee	ieee	NOUN
ajst-16801	206	2	trans	trans	PROPN
ajst-16801	206	3	pattern	pattern	PROPN
ajst-16801	206	4	anal	anal	ADJ
ajst-16801	206	5	mach	mach	NOUN
ajst-16801	206	6	intell	intell	NOUN
ajst-16801	206	7	31(2):210–227	31(2):210–227	PROPN
ajst-16801	206	8	.	.	PUNCT
ajst-16801	207	1	[	[	X
ajst-16801	207	2	20	20	NUM
ajst-16801	207	3	]	]	PUNCT
ajst-16801	207	4	elhamifar	elhamifar	PROPN
ajst-16801	207	5	e	e	NOUN
ajst-16801	207	6	,	,	PUNCT
ajst-16801	207	7	vidal	vidal	PROPN
ajst-16801	207	8	r	r	NOUN
ajst-16801	207	9	(	(	PUNCT
ajst-16801	207	10	2011	2011	NUM
ajst-16801	207	11	)	)	PUNCT
ajst-16801	207	12	robust	robust	ADJ
ajst-16801	207	13	classification	classification	NOUN
ajst-16801	207	14	using	use	VERB
ajst-16801	207	15	structured	structure	VERB
ajst-16801	207	16	sparse	sparse	ADJ
ajst-16801	207	17	representation	representation	NOUN
ajst-16801	207	18	.	.	PUNCT
ajst-16801	208	1	[	[	X
ajst-16801	208	2	21	21	NUM
ajst-16801	208	3	]	]	X
ajst-16801	208	4	michael	michael	PROPN
ajst-16801	208	5	g	g	PROPN
ajst-16801	208	6	,	,	PUNCT
ajst-16801	208	7	stephen	stephen	PROPN
ajst-16801	208	8	b	b	PROPN
ajst-16801	208	9	(	(	PUNCT
ajst-16801	208	10	2013	2013	NUM
ajst-16801	208	11	)	)	PUNCT
ajst-16801	208	12	cvx	cvx	PROPN
ajst-16801	208	13	:	:	PUNCT
ajst-16801	208	14	matlab	matlab	PROPN
ajst-16801	208	15	software	software	NOUN
ajst-16801	208	16	for	for	ADP
ajst-16801	208	17	disciplined	discipline	VERB
ajst-16801	208	18	convex	convex	NOUN
ajst-16801	208	19	programming	programming	NOUN
ajst-16801	208	20	,	,	PUNCT
ajst-16801	208	21	version	version	NOUN
ajst-16801	208	22	2.0	2.0	NUM
ajst-16801	208	23	beta	beta	NOUN
ajst-16801	208	24	.	.	PUNCT
ajst-16801	209	1	[	[	X
ajst-16801	209	2	22	22	NUM
ajst-16801	209	3	]	]	X
ajst-16801	209	4	zhang	zhang	PROPN
ajst-16801	209	5	,	,	PUNCT
ajst-16801	209	6	l	l	PROPN
ajst-16801	209	7	,	,	PUNCT
ajst-16801	209	8	yang	yang	PROPN
ajst-16801	209	9	m	m	PROPN
ajst-16801	209	10	,	,	PUNCT
ajst-16801	209	11	feng	feng	PROPN
ajst-16801	209	12	x	x	X
ajst-16801	209	13	(	(	PUNCT
ajst-16801	209	14	2011	2011	NUM
ajst-16801	209	15	)	)	PUNCT
ajst-16801	209	16	sparse	sparse	ADJ
ajst-16801	209	17	representation	representation	NOUN
ajst-16801	209	18	or	or	CCONJ
ajst-16801	209	19	collaborative	collaborative	ADJ
ajst-16801	209	20	representation	representation	NOUN
ajst-16801	209	21	:	:	PUNCT
ajst-16801	209	22	which	which	PRON
ajst-16801	209	23	helps	help	VERB
ajst-16801	209	24	face	face	VERB
ajst-16801	209	25	recognition	recognition	NOUN
ajst-16801	209	26	.	.	PUNCT
ajst-16801	210	1	[	[	X
ajst-16801	210	2	23	23	NUM
ajst-16801	210	3	]	]	X
ajst-16801	210	4	jiang	jiang	PROPN
ajst-16801	210	5	z	z	PROPN
ajst-16801	210	6	,	,	PUNCT
ajst-16801	210	7	lin	lin	PROPN
ajst-16801	210	8	z	z	PROPN
ajst-16801	210	9	,	,	PUNCT
ajst-16801	210	10	davis	davis	PROPN
ajst-16801	210	11	ls	ls	PROPN
ajst-16801	210	12	(	(	PUNCT
ajst-16801	210	13	2013	2013	NUM
ajst-16801	210	14	)	)	PUNCT
ajst-16801	210	15	label	label	NOUN
ajst-16801	210	16	consistent	consistent	ADJ
ajst-16801	210	17	k	k	PROPN
ajst-16801	210	18	-	-	PUNCT
ajst-16801	210	19	svd	svd	PROPN
ajst-16801	210	20	:	:	PUNCT
ajst-16801	210	21	learning	learn	VERB
ajst-16801	210	22	a	a	DET
ajst-16801	210	23	discriminative	discriminative	NOUN
ajst-16801	210	24	dictionary	dictionary	NOUN
ajst-16801	210	25	for	for	ADP
ajst-16801	210	26	recognition	recognition	NOUN
ajst-16801	210	27	.	.	PUNCT
ajst-16801	211	1	ieee	ieee	PROPN
ajst-16801	211	2	trans	trans	PROPN
ajst-16801	211	3	pattern	pattern	PROPN
ajst-16801	211	4	anal	anal	PROPN
ajst-16801	211	5	mach	mach	PROPN
ajst-16801	211	6	intell	intell	PROPN
ajst-16801	211	7	35(11):2651–2664	35(11):2651–2664	NUM
ajst-16801	211	8	.	.	PUNCT
ajst-16801	212	1	[	[	X
ajst-16801	212	2	24	24	NUM
ajst-16801	212	3	]	]	X
ajst-16801	212	4	yang	yang	PROPN
ajst-16801	212	5	m	m	PROPN
ajst-16801	212	6	,	,	PUNCT
ajst-16801	212	7	zhang	zhang	PROPN
ajst-16801	212	8	l	l	PROPN
ajst-16801	212	9	,	,	PUNCT
ajst-16801	212	10	feng	feng	PROPN
ajst-16801	212	11	x	x	SYM
ajst-16801	212	12	,	,	PUNCT
ajst-16801	212	13	zhang	zhang	PROPN
ajst-16801	212	14	d	d	X
ajst-16801	212	15	(	(	PUNCT
ajst-16801	212	16	2011	2011	NUM
ajst-16801	212	17	)	)	PUNCT
ajst-16801	212	18	fisher	fisher	PROPN
ajst-16801	212	19	discrimination	discrimination	NOUN
ajst-16801	212	20	dictionary	dictionary	NOUN
ajst-16801	212	21	learning	learn	VERB
ajst-16801	212	22	for	for	ADP
ajst-16801	212	23	sparse	sparse	ADJ
ajst-16801	212	24	representation	representation	NOUN
ajst-16801	212	25	.	.	PUNCT
ajst-16801	213	1	[	[	X
ajst-16801	213	2	25	25	NUM
ajst-16801	213	3	]	]	X
ajst-16801	213	4	cai	cai	X
ajst-16801	213	5	s	s	PROPN
ajst-16801	213	6	,	,	PUNCT
ajst-16801	213	7	zuo	zuo	PROPN
ajst-16801	213	8	w	w	PROPN
ajst-16801	213	9	,	,	PUNCT
ajst-16801	213	10	zhang	zhang	PROPN
ajst-16801	213	11	l	l	PROPN
ajst-16801	213	12	,	,	PUNCT
ajst-16801	213	13	feng	feng	PROPN
ajst-16801	213	14	x	x	PROPN
ajst-16801	213	15	,	,	PUNCT
ajst-16801	213	16	wang	wang	PROPN
ajst-16801	213	17	p	p	PROPN
ajst-16801	213	18	(	(	PUNCT
ajst-16801	213	19	2014	2014	NUM
ajst-16801	213	20	)	)	PUNCT
ajst-16801	213	21	support	support	NOUN
ajst-16801	213	22	vector	vector	NOUN
ajst-16801	213	23	guided	guide	VERB
ajst-16801	213	24	dictionary	dictionary	ADJ
ajst-16801	213	25	learning	learning	NOUN
ajst-16801	213	26	.	.	PUNCT
ajst-16801	214	1	[	[	X
ajst-16801	214	2	26	26	NUM
ajst-16801	214	3	]	]	X
ajst-16801	214	4	gu	gu	NOUN
ajst-16801	214	5	s	s	PROPN
ajst-16801	214	6	,	,	PUNCT
ajst-16801	214	7	zhang	zhang	PROPN
ajst-16801	214	8	l	l	PROPN
ajst-16801	214	9	,	,	PUNCT
ajst-16801	214	10	zuo	zuo	PROPN
ajst-16801	214	11	w	w	PROPN
ajst-16801	214	12	,	,	PUNCT
ajst-16801	214	13	feng	feng	PROPN
ajst-16801	214	14	x	x	X
ajst-16801	214	15	(	(	PUNCT
ajst-16801	214	16	2014	2014	NUM
ajst-16801	214	17	)	)	PUNCT
ajst-16801	214	18	projective	projective	ADJ
ajst-16801	214	19	dictionary	dictionary	ADJ
ajst-16801	214	20	pair	pair	NOUN
ajst-16801	214	21	learning	learn	VERB
ajst-16801	214	22	for	for	ADP
ajst-16801	214	23	pattern	pattern	NOUN
ajst-16801	214	24	classification	classification	NOUN
ajst-16801	214	25	.	.	PUNCT
ajst-16801	215	1	[	[	X
ajst-16801	215	2	27	27	NUM
ajst-16801	215	3	]	]	X
ajst-16801	215	4	wang	wang	PROPN
ajst-16801	215	5	y	y	PROPN
ajst-16801	215	6	,	,	PUNCT
ajst-16801	215	7	du	du	PROPN
ajst-16801	215	8	h	h	PROPN
ajst-16801	215	9	,	,	PUNCT
ajst-16801	215	10	zhang	zhang	PROPN
ajst-16801	215	11	y	y	PROPN
ajst-16801	215	12	,	,	PUNCT
ajst-16801	215	13	zhang	zhang	PROPN
ajst-16801	215	14	y	y	PROPN
ajst-16801	215	15	(	(	PUNCT
ajst-16801	215	16	2021	2021	NUM
ajst-16801	215	17	)	)	PUNCT
ajst-16801	215	18	efficient	efficient	ADJ
ajst-16801	215	19	and	and	CCONJ
ajst-16801	215	20	robust	robust	ADJ
ajst-16801	215	21	discriminant	discriminant	NOUN
ajst-16801	215	22	dictionary	dictionary	ADJ
ajst-16801	215	23	pair	pair	NOUN
ajst-16801	215	24	learning	learn	VERB
ajst-16801	215	25	for	for	ADP
ajst-16801	215	26	pattern	pattern	NOUN
ajst-16801	215	27	classification	classification	NOUN
ajst-16801	215	28	.	.	PUNCT
ajst-16801	216	1	digital	digital	ADJ
ajst-16801	216	2	signal	signal	NOUN
ajst-16801	216	3	process	process	NOUN
ajst-16801	216	4	118:103227	118:103227	NUM
ajst-16801	216	5	.	.	PUNCT
ajst-16801	217	1	[	[	X
ajst-16801	217	2	28	28	NUM
ajst-16801	217	3	]	]	X
ajst-16801	217	4	zhang	zhang	PROPN
ajst-16801	217	5	z	z	PROPN
ajst-16801	217	6	et	et	PROPN
ajst-16801	217	7	al	al	PROPN
ajst-16801	217	8	(	(	PUNCT
ajst-16801	217	9	2021	2021	NUM
ajst-16801	217	10	)	)	PUNCT
ajst-16801	217	11	twin	twin	ADJ
ajst-16801	217	12	-	-	PUNCT
ajst-16801	217	13	incoherent	incoherent	ADJ
ajst-16801	217	14	self	self	NOUN
ajst-16801	217	15	-	-	PUNCT
ajst-16801	217	16	expressive	expressive	ADJ
ajst-16801	217	17	localityadaptive	localityadaptive	ADJ
ajst-16801	217	18	latent	latent	NOUN
ajst-16801	217	19	dictionary	dictionary	ADJ
ajst-16801	217	20	pair	pair	NOUN
ajst-16801	217	21	learning	learn	VERB
ajst-16801	217	22	for	for	ADP
ajst-16801	217	23	classification	classification	NOUN
ajst-16801	217	24	.	.	PUNCT
ajst-16801	218	1	ieee	ieee	PROPN
ajst-16801	218	2	trans	trans	PROPN
ajst-16801	218	3	neural	neural	PROPN
ajst-16801	218	4	netw	netw	NOUN
ajst-16801	218	5	learn	learn	VERB
ajst-16801	218	6	syst	syst	PROPN
ajst-16801	218	7	323:947	323:947	NOUN
ajst-16801	218	8	–	–	PUNCT
ajst-16801	218	9	961	961	NUM
ajst-16801	218	10	.	.	PUNCT
ajst-16801	219	1	[	[	X
ajst-16801	219	2	29	29	NUM
ajst-16801	219	3	]	]	X
ajst-16801	219	4	dong	dong	PROPN
ajst-16801	219	5	j	j	PROPN
ajst-16801	219	6	,	,	PUNCT
ajst-16801	219	7	yang	yang	PROPN
ajst-16801	219	8	l	l	PROPN
ajst-16801	219	9	,	,	PUNCT
ajst-16801	219	10	liu	liu	PROPN
ajst-16801	219	11	c	c	PROPN
ajst-16801	219	12	,	,	PUNCT
ajst-16801	219	13	cheng	cheng	PROPN
ajst-16801	219	14	w	w	PROPN
ajst-16801	219	15	,	,	PUNCT
ajst-16801	219	16	wang	wang	PROPN
ajst-16801	219	17	w	w	PROPN
ajst-16801	219	18	(	(	PUNCT
ajst-16801	219	19	2022	2022	NUM
ajst-16801	219	20	)	)	PUNCT
ajst-16801	219	21	support	support	NOUN
ajst-16801	219	22	vector	vector	NOUN
ajst-16801	219	23	machine	machine	NOUN
ajst-16801	219	24	embedding	embed	VERB
ajst-16801	219	25	discriminative	discriminative	NOUN
ajst-16801	219	26	dictionary	dictionary	ADJ
ajst-16801	219	27	pair	pair	NOUN
ajst-16801	219	28	learning	learn	VERB
ajst-16801	219	29	for	for	ADP
ajst-16801	219	30	pattern	pattern	NOUN
ajst-16801	219	31	classification	classification	NOUN
ajst-16801	219	32	.	.	PUNCT
ajst-16801	220	1	neural	neural	ADJ
ajst-16801	220	2	netw	netw	PROPN
ajst-16801	220	3	155:498–511	155:498–511	NUM
ajst-16801	220	4	.	.	PUNCT
ajst-16801	221	1	[	[	X
ajst-16801	221	2	30	30	NUM
ajst-16801	221	3	]	]	X
ajst-16801	221	4	ji	ji	PROPN
ajst-16801	221	5	p	p	PROPN
ajst-16801	221	6	,	,	PUNCT
ajst-16801	221	7	zhang	zhang	PROPN
ajst-16801	221	8	t	t	PROPN
ajst-16801	221	9	,	,	PUNCT
ajst-16801	221	10	li	li	PROPN
ajst-16801	221	11	h	h	PROPN
ajst-16801	221	12	,	,	PUNCT
ajst-16801	221	13	salzmann	salzmann	PROPN
ajst-16801	221	14	m	m	PROPN
ajst-16801	221	15	,	,	PUNCT
ajst-16801	221	16	reid	reid	PROPN
ajst-16801	221	17	i	i	PROPN
ajst-16801	221	18	(	(	PUNCT
ajst-16801	221	19	2017	2017	NUM
ajst-16801	221	20	)	)	PUNCT
ajst-16801	221	21	deep	deep	ADJ
ajst-16801	221	22	subspace	subspace	NOUN
ajst-16801	221	23	clustering	cluster	VERB
ajst-16801	221	24	networks	network	NOUN
ajst-16801	221	25	.	.	PUNCT
ajst-16801	222	1	[	[	X
ajst-16801	222	2	31	31	NUM
ajst-16801	222	3	]	]	X
ajst-16801	222	4	zhang	zhang	PROPN
ajst-16801	222	5	j	j	PROPN
ajst-16801	222	6	,	,	PUNCT
ajst-16801	222	7	et	et	PROPN
ajst-16801	222	8	al	al	PROPN
ajst-16801	222	9	.	.	PROPN
ajst-16801	222	10	(	(	PUNCT
ajst-16801	222	11	2019	2019	NUM
ajst-16801	222	12	)	)	PUNCT
ajst-16801	222	13	self	self	NOUN
ajst-16801	222	14	-	-	PUNCT
ajst-16801	222	15	supervised	supervise	VERB
ajst-16801	222	16	convolutional	convolutional	ADJ
ajst-16801	222	17	subspace	subspace	NOUN
ajst-16801	222	18	clustering	cluster	VERB
ajst-16801	222	19	network	network	NOUN
ajst-16801	222	20	,	,	PUNCT
ajst-16801	222	21	5468	5468	NUM
ajst-16801	222	22	-	-	SYM
ajst-16801	222	23	5477	5477	NUM
ajst-16801	222	24	.	.	PUNCT