Reproduce “Align Before You Combine: Reference Space Calibration for Supervision Without Ground Truth” 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
Jackson Eshbaugh, Jorge Silveyra
The authors introduce calibration-first fusion to combine scores from sources with incomparable scales. They score synthetic reference examples, standardize each scorer against its reference mean and standard deviation, then average aligned scores. Across three benchmarks, it beat raw averaging on primary metrics; on Wine Quality, Spearman correlation was 0.556 versus 0.192. This shows rank correspondence, not universal validity or downstream improvement.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.