fcParameterBackwardPass: Backpropagation (parameters' deltas) for fully connected...

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

Description

This function calculates parameters' deltas at a given layer when using more than one observation at the time.

Usage

fcParameterBackwardPass(
  deltaMw,
  deltaSw,
  deltaMb,
  deltaSb,
  Sw,
  Sb,
  ma,
  deltaMr,
  deltaSr,
  ni,
  no,
  B,
  rB
)

Arguments

deltaMw	next layer delta of mean vector of weights given y μ_{θ}|y
deltaSw	next layer delta of covariance matrix of weights given y Σ_{θ}|y
deltaMb	next layer delta of mean vector of biases given y μ_{θ}|y
deltaSb	next layer delta of covariance matrix of biases given y Σ_{θ}|y
Sw	Covariance of weights for the current layer
Sb	Covariance of biases for the current layer
ma	Mean vector of activation units for the current layer
deltaMr	Delta of mean vector of next layer units given y μ_{Z}|y
deltaSr	Delta of covariance matrix of next layer units given y Σ_{Z}|y
ni	Number of units in current layer
no	Number of units in next layer
B	Batch size
rB	Number of times batch size is repeated

Value

• Delta of mean vector of weights given y μ_{θ}|y

• Delta of covariance matrix of weights given y Σ_{θ}|y

• Delta of mean vector of biases given y μ_{θ}|y

• Delta of covariance matrix of biases given y Σ_{θ}|y

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