Reproduce “Conformal Prediction Sets Quantify Information Gain: A Theoretical Perspective” 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
Kevin Zhang, Stephen Bates
The authors define a family of information measures for conformal set-valued predictions, with Shannon mutual information equal to its integral and set-size reduction bounded by measures near calibrated thresholds, subject to calibration and model slack. On CIFAR-100 at 90% nominal coverage, set-size-based selection averaged 12.683 versus 13.763 for Shannon-MI selection, across global feature orderings, not per-example acquisition.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.