Explanations about an action (human interpretability)

[luang008]

Very cool project. I know it's a hard problem, but just wondering if there are related efforts to not only show the model numeric scores, but also a human understandable reasoning, similar to basic “何切” discussions or even ML pro “复盘”.

Context: I am interested on exploring it and thus want to know what has been tried.

Thanks

[Equim-chan]

I experimented with some basic explainable AI methods like saliency maps, but they were ineffective.

[luang008]

Thanks for the quick response. Let me play around and get back with any findings.