fcDerivative2: Second derivatives for fully connected layers

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

Description

This function calculates mean of product of derivatives, when new product term involves second derivatives (wdd).

Usage

fcDerivative2(
  mw,
  mwo,
  mao,
  mai,
  mdai,
  mddai,
  mpddi,
  mdgo,
  mdgo2,
  Caoai,
  Cdow,
  Cdodi,
  acti,
  ni,
  no,
  no2,
  B
)

Arguments

mw	Mean vector of weights for the current layer
mwo	Mean vector of weights for the next layer
mao	Mean vector of activation units from next layer
mai	Mean vector of activation units from current layer
mdai	Mean vector of activation units' first derivative from current layer
mddai	Mean vector of activation units' second derivative from current layer
mpddi	Mean vector of second derivative product wdd of current layer
mdgo	Mean vector of product of derivatives in next layer
mdgo2	Mean vector of product of derivatives in second next layer
Caoai	Covariance between activation units from current and next layers
Cdow	Covariance between first derivatives and weights
Cdodi	Covariance between first derivatives from current and next layers
acti	Activation function index for current layer defined by activationFunIndex
ni	Number of units in current layer
no	Number of units in next layer
no2	Number of units in second next layer
B	Batch size

Value

Mean of product terms for second derivative calculations

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