search

albermax / innvestigate

A toolbox to iNNvestigate neural networks' predictions!
Other
1.26k stars 233 forks source link

[REQUEST]How LRP is used for obtaining the relevance score of each neuron? #271

Closed TestingAIGroup closed 2 years ago

TestingAIGroup commented 2 years ago

Hello, Thank you very much for your work. I am really interested in your work. I would like to learn how LRP is used for obtaining the relevance score of each neuron? Could you please help me to solve this issue? Thank you very much.

nicrie commented 2 years ago

A good starting point could be some of the original references, e.g. Bach et al. 2015 or Montavon et al. 2019.

sebastian-lapuschkin commented 2 years ago

I think the question was more in context of innvestigate, ie "how to access latent representation relevances". I am calling @adrhill to the task.

adrhill commented 2 years ago

Hi @TestingAIGroup,

all LRP analyzers take the optional parameter reverse_keep_tensors=True, which will store all intermediate tensors on the backward pass.

An example is shown in the last cell of the notebook "Developing with iNNvestigate" and more details on optional keyword arguments can be found in the docs.