search

albermax / innvestigate

A toolbox to iNNvestigate neural networks' predictions!
Other
1.25k stars 233 forks source link

Composite LRP #162

Open enryH opened 4 years ago

enryH commented 4 years ago

Dear team,

if I would like to do composite LRP using iNNvestigate as described in the Figure 10.4 in your latest overview-paper:

Montavon G., Binder A., Lapuschkin S., Samek W., Müller KR. (2019) Layer-Wise Relevance Propagation: An Overview. In: Samek W., Montavon G., Vedaldi A., Hansen L., Muller KR. (eds) Explainable AI: Interpreting, Explaining and Visualizing Deep Learning. Lecture Notes in Computer Science, vol 11700. Springer, Cham

I am thinking about how to do this in iNNvestigate. I guess I could split the network into different parts, and do respective analysis on each of them and pass the relevances backwards. Is there an easier way?

Thanks in advance!

Argho-UMBC commented 4 years ago

Hey, did you find any solution for composite LRP?

sebastian-lapuschkin commented 3 years ago

you can use the lrp.sequential_* analyzers. see https://github.com/albermax/innvestigate/blob/c6ce89323d686a52d793177efd8d1dda3bd9526a/src/innvestigate/analyzer/__init__.py#L60

those are implemented by applying layer-conditional rule mapping.