Skip to content

Back propagation equations #5

@tessavdheiden

Description

@tessavdheiden

Hi,

Thanks for your code. It is very well structured.

Are you sure that the first three equations in the back propagation (237-239) are correct? I think that the derivatives of the weights (dSdwW) should be computed by the derivative of NetOutputAct and CECSquashAct.

Moreover, for the NetOutputAct derivative, did you forget to use the derivative of the softmax?

Best
Tessa

Metadata

Metadata

Assignees

Labels

Projects

No projects

Milestone

No milestone

Relationships

None yet

Development

No branches or pull requests

Issue actions