Reproduce “On the Relationship Between Explanation and Prediction: A Causal View” on OpenResearch

Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.

Universal app for Apple silicon and Intel. Requires macOS 11 or later.

About this paper

Amir-Hossein Karimi, Krikamol Muandet, Simon Kornblith, Bernhard Schölkopf, Been Kim

The authors assess whether saliency explanations track predictions as training hyperparameters vary. They compare treatment effects across 30,000 three-layer CNNs for each of four image datasets, using four saliency methods and an Identity reference. The methods do not match that idealized reference; the gap grows with model performance, within this model zoo and an approximate analysis.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.