search

giotto-ai / giotto-deep

Deep learning made topological.
Other
79 stars 11 forks source link

New method `get_activation_gradient` #71

Closed matteocao closed 7 months ago

matteocao commented 2 years ago

Is your feature request related to a problem? Please describe.

There is a need to build a method that extracts the gradients with respect to the activations, hence:

\frac{d(Loss)}{d (activation_i)}

Describe the solution you'd like

A possible solution would be to deepcopy the model, add biases, compute the gradient with respect to biases. Describe alternatives you've considered

Not sure, open to suggestions. Additional context

raphaelreinauer commented 2 years ago

Is this still relevant?

nberkouk commented 2 years ago

I tried to make it work and it did not. Then Matteo tried and it seemed a bit tricky. It is relevant at least to my own personal research on adversarial attacks, but maybe it does not need to be part of the library, or it can be developed later

RaphaelLilt commented 7 months ago

@matteocao Can we close this issue?