Reproduce “Selective Prediction Intervals: Jointly Learning to Predict, Bound, and Abstain” 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
Lidor Rubi, Gilad Katz
SelPI combines regression estimates, intervals, and learned abstention in one system. It uses two jointly trained gates: one detects likely interval misses, and another identifies poorly placed point estimates when intervals capture targets. Across 18 tabular datasets, its single-gate selector beats width thresholding on interval score, with median selection efficiency 0.87. This benchmark result does not provide a finite-sample coverage guarantee for accepted predictions, and gains are smaller on near-deterministic tasks.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.