The generic Deep Taylor Decomposition formula in the paper

hila-chefer / Transformer-Explainability

[CVPR 2021] Official PyTorch implementation for Transformer Interpretability Beyond Attention Visualization, a novel method to visualize classifications by Transformer based networks.

MIT License

1.75k stars 232 forks source link

The generic Deep Taylor Decomposition formula in the paper #60

Open a943862842 opened 1 year ago

a943862842 commented 1 year ago

The generic Deep Taylor Decomposition formula in the paper seems to be different as the formula in reference 27. The Deep Taylor Decomposition formula in Reference 27 requires selecting a root. Could you please show me how this formula was derived? Thank you! 1691080115266

Nanuion commented 8 months ago

Does anyone know how the pos and neg metrics mentioned in the article are implemented in code?

jykr commented 2 months ago

@hila-chefer @shirgur I have same questions- Could you please comment on these?