ReproBridge improves ML experiment reproducibility using refined artifacts
Original title: ReproBridge: Building Reproducibility into ML Research Artifacts
Refined artifacts raise reproduction success from 76.4% to 92.6% and cut API spending by 19.7% on 1029 tasks from 36 papers.
Reproducing ML experiments requires filling gaps in scattered procedures from papers, code, and artifacts. ReproBridge is an agent workflow that builds standardized refined artifacts including codebases, setups, launchers, and guides. Across four evaluated models on ReproBridge-Bench, these artifacts improve success rates and narrow model performance gaps.
- 1029 tasks from 387 experiments across 36 papers
- Mean success rises from 76.4% to 92.6%
- API spending reduces by 19.7% on average
- All models exceed 90% success with refined artifacts
- Removing code or environment components drops success
From the original article
1 min read at arxiv.orgReproducing machine learning experiments requires reconstructing procedures from information scattered across papers, released code, and supporting artifacts. For coding agents, reproduction success can therefore depend on their ability to fill in missing details, rather than on the released artifacts alone. We ask whether preserving recovered procedures and decisions improves subsequent independent reproduction.
By Yubo Wang · arxiv.org
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.