Reproduce “Distilling Calibration via Conformalized Credal Inference” 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

Jiayi Huang, Sangwoo Park, Nicola Paoletti, Osvaldo Simeone

CD-CI uses an offline cloud model to calibrate a small edge model, forming plausible prediction sets without the cloud at test time. Under exchangeability, these contain the cloud model’s predictive distribution at target coverage, not necessarily true labels. On CIFAR-10’s three-class task, ECE fell approximately 4% versus the small model and 3% versus Laplace at coverage 0.85 to 0.92, with no reported accuracy loss.

Summary by alphaXiv.

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