AI
AdaPRA: Process Reward Agents with Adaptive Rubrics for Data Science
Publication: cs.AI updates on arXiv.org
Author: Atharva Naik, Vivek Chari, Pol Llado, Daniel Fried, Carolyn Ros\'e, Benjamin Van Durme, Sam Thomson, Justin Svegliato
Source published: 27/10/2026, 04:00:00 UTC
Briefing created: 22/11/2026, 16:41:08 UTC
Based on a feed excerpt; full article content has not been acquired. Source evidence retrieved 28/10/2026, 02:33:25 UTC.
Briefings
AdaPRA is a persistent coding agent that uses rubric-guided process supervision to correct errors in data science workflows. It improves an 80B Qwen model's performance by 24.5 points on DSPredict-Easy and 14.8 points on DSPredict-Hard across five attempts per task, reducing process errors. The system adapts its rubrics to each task, showing that rubrics seeded from easier tasks generalize to harder tasks. AdaPRA also outperforms DataPRM, a competitive process reward model, and shares code for reproducibility.
Key points
- AdaPRA improves an 80B Qwen model's leaderboard percentile by 24.5 points on DSPredict-Easy and 14.8 points on DSPredict-Hard across five attempts per task.
- Rubrics seeded from easier tasks generalize to harder tasks, with rubric adaptation further improving performance.
Context
Data science agents automate end-to-end machine learning workflows but remain far from competitive on human leaderboards. Agents make process errors such as insufficient exploration, unsound validation, coding bugs, and poor use of a fixed budget. AdaPRA proposes a persistent coding agent harness that accumulates knowledge across attempts on the same task, with rubric-guided process supervision that identifies and corrects errors.
Limitations
Evidence is from a preprint (arXiv:2610.33245v1) and covers only five attempts per task. Performance improvements are reported on specific datasets (DSPredict-Easy/Hard) and models (80B Qwen, DeepSeek-V3.2). No additional background or context is supplied beyond the abstract.
Generated by Bittrees-hosted qwen35-2b sha256:aaf42c8b7c3c on 22/11/2026, 16:41:08 UTC. Based on publisher-provided feed evidence; may not cover the full original article. Check the source for details.
View archived briefing on IPFS
bafybeiddc7tbuahoaib3nsa5fwffgecf742n74mvzhrurefirlgtfmfy34