A Reweighted Scheme to Improve the Representation of the Neural Autoregressive Distribution Estimator

Table 8

Classification result for initialization on twelve datasets.

Dataset	Input size	Hidden layer size	NADE	Reweighted-NADE	RBM
Adult	123	100	0.15921	0.15814	0.16009
Binarized-MNIST	784	300	0.0216	0.0198	0.0205
Connect-4	126	100	0.19326	0.18474	0.19725
Convex	784	300	0.25672	0.23898	0.26394
DNA	126	100	0.06998	0.05986	0.06324
Mushrooms	112	100	0.00729	0.00498	0.00605
Newsgroups	5000	1000	0.34231	0.28767	0.3291
OCR-letters	128	100	0.1669	0.1601	0.1374
RCV1	150	100	0.05131	0.04449	0.05567
Rectangles	784	300	0.09062	0.08338	0.09948
SVHN	1024	300	0.08405	0.07810	0.08889
Web	300	150	0.01477	0.01366	0.01492