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

Algorithm 2

A reweighted scheme for initialization.

	Given: A set of training samples , weight matrix , and bias from the trained NADE
	Output: A reweighted new weight matrix and bias for pretraining the neural network
(1)	For training sample , compute the corresponding reweighted score vector using equations (13)–(16)
(2)	Using equation (20) to compute the final reweighted score vector
(3)	Using equations (21) and (22) to obtain and
(4)	Choose the best weight matrix and bias by validation.