Reproduce “Bayesian Importance of Features (BIF)” 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
Kamil Adamczewski, Frederik Harder, Mijung Park
BIF learns feature importance for a fixed model, producing global or example-specific relative weights and uncertainty. It uses a Dirichlet distribution, updating its parameters while leaving the model unchanged. On Syn4, a switching-feature task, instance-wise sampling reports MCC 81.9 versus L2X’s 66.0, measuring agreement between selected and relevant features. This supports feature identification on tested tasks, not calibrated uncertainty or causal importance.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.