A Critical Look At Some Common Trends In IML Research

Note: The event time listed is set to Pacific Time.

I wish to stimulate a critical discussion of some popular trends in IML publications. Topics I would like to address include:

1) Accuracy interpretability tradeoffs for ML class predictors.

2) Causality as interpretation and counterfactual analysis as causality.

3) Heat maps (and saliency masks) as interpretations for DNNs.

4) Interpreting image-interpretations beyond object recognition.