Reproduce “Benchmarking uncertainty quantification for engineering regression under graded distribution shift” 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
Ahmad Lawal, Rahim Pasbanigoloojeh, Asmau Lawal Bilbis
The authors evaluate prediction intervals across seven engineering datasets using three conditions: in-distribution data, overlapping soft shifts, and hard shifts. They tune 12 methods using source data only, covering extrapolation and regime change. NGBoost’s coverage at a 95% target falls from 0.88 in-distribution to 0.40 under hard shift; 11 always-answering methods range from 0.40 to 0.75, with no difference detected.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.