Implement differentiable weight masks for modularity detection (from Are Neural Nets Modular? Inspecting Functional Modularity Through Differentiable Weight Masks)

[rteehas]

Using Csordás et al 2021 as a guide, implement differentiable weight masks to be used to detect modular substructures

[oserikov]

it seems like the paper implementation is available here, so the actual todo is:

• add the https://github.com/RobertCsordas/modules as a submodule in our rep
• Implement the ability to do smth like import diff_w_masks_probing; diff_w_masks_probing(huggingface_model_name, path_to_data)