ddotprod (Neural Network Toolbox)

Dot product weight derivative function

dZ_dP = ddotprod('p',W,P,Z)

dZ_dW = ddotprod('w',W,P,Z)

ddotprod is a weight derivative function.

ddotprod('p',W,P,Z) takes three arguments,

W - S x R weight matrix.

P - R x Q Inputs.

Z - S x Q weighted input.

and returns the S x R derivative dZ/dP.

ddotprod('w',W,P,Z) returns the R x Q derivative dZ/dW.

Examples

Here we define a weight W and input P for an input with 3 elements and a layer with 2 neurons.

W = [0 -1 0.2; -1.1 1 0];
P = [0.1; 0.6; -0.2];

Here we calculate the weighted input with dotprod, then calculate each derivative with ddotprod.

Z = dotprod(W,P)
dZ_dP = ddotprod('p',W,P,Z)
dZ_dW = ddotprod('w',W,P,Z)

The derivative of a product of two elements with respect to one element is the other element.

dZ/dP = W
dZ/dW = P