id	sid	tid	token	lemma	pos
h128nc61d17	1	1	machine	machine	NOUN
h128nc61d17	1	2	learning	learning	NOUN
h128nc61d17	1	3	models	model	NOUN
h128nc61d17	1	4	struggle	struggle	VERB
h128nc61d17	1	5	to	to	PART
h128nc61d17	1	6	generalize	generalize	VERB
h128nc61d17	1	7	when	when	SCONJ
h128nc61d17	1	8	the	the	DET
h128nc61d17	1	9	number	number	NOUN
h128nc61d17	1	10	of	of	ADP
h128nc61d17	1	11	class	class	NOUN
h128nc61d17	1	12	instances	instance	NOUN
h128nc61d17	1	13	is	be	AUX
h128nc61d17	1	14	numerically	numerically	ADV
h128nc61d17	1	15	imbalanced	imbalanced	ADJ
h128nc61d17	1	16	.	.	PUNCT
h128nc61d17	2	1	data	datum	NOUN
h128nc61d17	2	2	augmentation	augmentation	NOUN
h128nc61d17	2	3	(	(	PUNCT
h128nc61d17	2	4	da	da	PROPN
h128nc61d17	2	5	)	)	PUNCT
h128nc61d17	2	6	is	be	AUX
h128nc61d17	2	7	a	a	DET
h128nc61d17	2	8	leading	lead	VERB
h128nc61d17	2	9	approach	approach	NOUN
h128nc61d17	2	10	to	to	PART
h128nc61d17	2	11	improve	improve	VERB
h128nc61d17	2	12	generalization	generalization	NOUN
h128nc61d17	2	13	for	for	ADP
h128nc61d17	2	14	under	under	ADV
h128nc61d17	2	15	-	-	PUNCT
h128nc61d17	2	16	represented	represent	VERB
h128nc61d17	2	17	classes	class	NOUN
h128nc61d17	2	18	.	.	PUNCT
h128nc61d17	3	1	despite	despite	SCONJ
h128nc61d17	3	2	its	its	PRON
h128nc61d17	3	3	wide	wide	ADV
h128nc61d17	3	4	-	-	PUNCT
h128nc61d17	3	5	spread	spread	NOUN
h128nc61d17	3	6	use	use	NOUN
h128nc61d17	3	7	,	,	PUNCT
h128nc61d17	3	8	the	the	DET
h128nc61d17	3	9	mechanisms	mechanism	NOUN
h128nc61d17	3	10	by	by	ADP
h128nc61d17	3	11	which	which	PRON
h128nc61d17	3	12	da	da	PROPN
h128nc61d17	3	13	works	work	NOUN
h128nc61d17	3	14	are	be	AUX
h128nc61d17	3	15	not	not	PART
h128nc61d17	3	16	clearly	clearly	ADV
h128nc61d17	3	17	understood.in	understood.in	NUM
h128nc61d17	3	18	this	this	DET
h128nc61d17	3	19	dissertation	dissertation	NOUN
h128nc61d17	3	20	,	,	PUNCT
h128nc61d17	3	21	we	we	PRON
h128nc61d17	3	22	take	take	VERB
h128nc61d17	3	23	a	a	DET
h128nc61d17	3	24	step	step	NOUN
h128nc61d17	3	25	toward	toward	ADP
h128nc61d17	3	26	understanding	understand	VERB
h128nc61d17	3	27	how	how	SCONJ
h128nc61d17	3	28	da	da	NOUN
h128nc61d17	3	29	works	work	VERB
h128nc61d17	3	30	with	with	ADP
h128nc61d17	3	31	imbalanced	imbalanced	ADJ
h128nc61d17	3	32	data	datum	NOUN
h128nc61d17	3	33	.	.	PUNCT
h128nc61d17	4	1	we	we	PRON
h128nc61d17	4	2	begin	begin	VERB
h128nc61d17	4	3	by	by	ADP
h128nc61d17	4	4	building	build	VERB
h128nc61d17	4	5	three	three	NUM
h128nc61d17	4	6	novel	novel	NOUN
h128nc61d17	4	7	algorithms	algorithm	NOUN
h128nc61d17	4	8	,	,	PUNCT
h128nc61d17	4	9	which	which	PRON
h128nc61d17	4	10	incorporate	incorporate	VERB
h128nc61d17	4	11	data	datum	NOUN
h128nc61d17	4	12	augmentation	augmentation	NOUN
h128nc61d17	4	13	,	,	PUNCT
h128nc61d17	4	14	to	to	PART
h128nc61d17	4	15	improve	improve	VERB
h128nc61d17	4	16	generalization	generalization	NOUN
h128nc61d17	4	17	for	for	ADP
h128nc61d17	4	18	under	under	ADV
h128nc61d17	4	19	-	-	PUNCT
h128nc61d17	4	20	represented	represent	VERB
h128nc61d17	4	21	classes	class	NOUN
h128nc61d17	4	22	.	.	PUNCT
h128nc61d17	5	1	based	base	VERB
h128nc61d17	5	2	on	on	ADP
h128nc61d17	5	3	insights	insight	NOUN
h128nc61d17	5	4	gleaned	glean	VERB
h128nc61d17	5	5	from	from	ADP
h128nc61d17	5	6	this	this	DET
h128nc61d17	5	7	process	process	NOUN
h128nc61d17	5	8	,	,	PUNCT
h128nc61d17	5	9	we	we	PRON
h128nc61d17	5	10	focus	focus	VERB
h128nc61d17	5	11	on	on	ADP
h128nc61d17	5	12	the	the	DET
h128nc61d17	5	13	latent	latent	ADJ
h128nc61d17	5	14	features	feature	NOUN
h128nc61d17	5	15	learned	learn	VERB
h128nc61d17	5	16	by	by	ADP
h128nc61d17	5	17	machine	machine	NOUN
h128nc61d17	5	18	learning	learning	NOUN
h128nc61d17	5	19	(	(	PUNCT
h128nc61d17	5	20	ml	ml	PROPN
h128nc61d17	5	21	)	)	PUNCT
h128nc61d17	5	22	models	model	NOUN
h128nc61d17	5	23	as	as	ADP
h128nc61d17	5	24	potential	potential	ADJ
h128nc61d17	5	25	culprits	culprit	NOUN
h128nc61d17	5	26	in	in	ADP
h128nc61d17	5	27	generalization	generalization	NOUN
h128nc61d17	5	28	.	.	PUNCT
h128nc61d17	6	1	we	we	PRON
h128nc61d17	6	2	design	design	VERB
h128nc61d17	6	3	a	a	DET
h128nc61d17	6	4	suite	suite	NOUN
h128nc61d17	6	5	of	of	ADP
h128nc61d17	6	6	tools	tool	NOUN
h128nc61d17	6	7	,	,	PUNCT
h128nc61d17	6	8	with	with	ADP
h128nc61d17	6	9	latent	latent	ADJ
h128nc61d17	6	10	features	feature	NOUN
h128nc61d17	6	11	,	,	PUNCT
h128nc61d17	6	12	that	that	PRON
h128nc61d17	6	13	can	can	AUX
h128nc61d17	6	14	be	be	AUX
h128nc61d17	6	15	used	use	VERB
h128nc61d17	6	16	to	to	PART
h128nc61d17	6	17	understand	understand	VERB
h128nc61d17	6	18	data	data	NOUN
h128nc61d17	6	19	complexity	complexity	NOUN
h128nc61d17	6	20	and	and	CCONJ
h128nc61d17	6	21	class	class	NOUN
h128nc61d17	6	22	overlap.we	overlap.we	NUM
h128nc61d17	6	23	also	also	ADV
h128nc61d17	6	24	find	find	VERB
h128nc61d17	6	25	that	that	SCONJ
h128nc61d17	6	26	certain	certain	ADJ
h128nc61d17	6	27	da	da	NOUN
h128nc61d17	6	28	methods	method	NOUN
h128nc61d17	6	29	and	and	CCONJ
h128nc61d17	6	30	parametric	parametric	NOUN
h128nc61d17	6	31	ml	ml	PROPN
h128nc61d17	6	32	classifiers	classifiers	PROPN
h128nc61d17	6	33	(	(	PUNCT
h128nc61d17	6	34	cnn	cnn	PROPN
h128nc61d17	6	35	,	,	PUNCT
h128nc61d17	6	36	logistic	logistic	ADJ
h128nc61d17	6	37	regression	regression	NOUN
h128nc61d17	6	38	,	,	PUNCT
h128nc61d17	6	39	svm	svm	PROPN
h128nc61d17	6	40	)	)	PUNCT
h128nc61d17	6	41	incorporate	incorporate	VERB
h128nc61d17	6	42	hidden	hidden	ADJ
h128nc61d17	6	43	linearity	linearity	NOUN
h128nc61d17	6	44	at	at	ADP
h128nc61d17	6	45	the	the	DET
h128nc61d17	6	46	front	front	ADJ
h128nc61d17	6	47	-	-	PUNCT
h128nc61d17	6	48	end	end	NOUN
h128nc61d17	6	49	of	of	ADP
h128nc61d17	6	50	training	training	NOUN
h128nc61d17	6	51	and	and	CCONJ
h128nc61d17	6	52	during	during	ADP
h128nc61d17	6	53	inference	inference	NOUN
h128nc61d17	6	54	that	that	PRON
h128nc61d17	6	55	may	may	AUX
h128nc61d17	6	56	affect	affect	VERB
h128nc61d17	6	57	generalization	generalization	NOUN
h128nc61d17	6	58	,	,	PUNCT
h128nc61d17	6	59	when	when	SCONJ
h128nc61d17	6	60	learning	learn	VERB
h128nc61d17	6	61	with	with	ADP
h128nc61d17	6	62	imbalanced	imbalanced	ADJ
h128nc61d17	6	63	data	datum	NOUN
h128nc61d17	6	64	.	.	PUNCT
h128nc61d17	7	1	further	far	ADV
h128nc61d17	7	2	,	,	PUNCT
h128nc61d17	7	3	we	we	PRON
h128nc61d17	7	4	demonstrate	demonstrate	VERB
h128nc61d17	7	5	that	that	SCONJ
h128nc61d17	7	6	parametric	parametric	NOUN
h128nc61d17	7	7	ml	ml	NOUN
h128nc61d17	7	8	models	model	NOUN
h128nc61d17	7	9	rely	rely	VERB
h128nc61d17	7	10	heavily	heavily	ADV
h128nc61d17	7	11	on	on	ADP
h128nc61d17	7	12	the	the	DET
h128nc61d17	7	13	magnitude	magnitude	NOUN
h128nc61d17	7	14	of	of	ADP
h128nc61d17	7	15	a	a	DET
h128nc61d17	7	16	limited	limited	ADJ
h128nc61d17	7	17	number	number	NOUN
h128nc61d17	7	18	of	of	ADP
h128nc61d17	7	19	latent	latent	ADJ
h128nc61d17	7	20	features	feature	NOUN
h128nc61d17	7	21	.	.	PUNCT
h128nc61d17	8	1	during	during	ADP
h128nc61d17	8	2	inference	inference	NOUN
h128nc61d17	8	3	,	,	PUNCT
h128nc61d17	8	4	they	they	PRON
h128nc61d17	8	5	predict	predict	VERB
h128nc61d17	8	6	classes	class	NOUN
h128nc61d17	8	7	based	base	VERB
h128nc61d17	8	8	on	on	ADP
h128nc61d17	8	9	a	a	DET
h128nc61d17	8	10	combination	combination	NOUN
h128nc61d17	8	11	of	of	ADP
h128nc61d17	8	12	latent	latent	ADJ
h128nc61d17	8	13	feature	feature	NOUN
h128nc61d17	8	14	magnitudes	magnitude	NOUN
h128nc61d17	8	15	that	that	PRON
h128nc61d17	8	16	sum	sum	VERB
h128nc61d17	8	17	to	to	ADP
h128nc61d17	8	18	a	a	DET
h128nc61d17	8	19	requisite	requisite	ADJ
h128nc61d17	8	20	threshold	threshold	NOUN
h128nc61d17	8	21	.	.	PUNCT