fcMeanDlayer2array: Mean of weights times derivatives products terms ((wdo*wdo) x...

In mgoulet847/tagi: Tractable Approximate Gaussian Inference in Neural Networks

Description

This function calculates mean of weights times derivatives products terms when adding two of those products from current layer to already calculated expectation that ended with one such product of next layer (i.e. (wdowdo) x (wwdi^2)). Mean terms are in array format. Once added, rows need to be summed to aggregate expectations by node*weight combinations of current layer.

Usage

fcMeanDlayer2array(mpdi2w, mdgo, Cwdowdowwdi2, ni, no, B)

Arguments

mpdi2w	Combination of products of first derivative of current layer (wd)*(wd) (iterations on weights on the same node)
mdgo	Mean of product of derivatives in next layer
Cwdowdowwdi2	Covariance cov(wdowdo,wdiwdi) of weights times derivatives products terms, where the di terms are the same
ni	Number of units in current layer
no	Number of units in next layer
B	Batch size

Value

Mean of weights times derivatives products terms

mgoulet847/tagi documentation built on Dec. 21, 2021, 5:10 p.m.