Learning to Reason with Curriculum I: Provable Benefits of Autocurriculum

Nived Rajaraman, Audrey Huang, Miro Dudik, Robert Schapire, Dylan J. Foster, Akshay Krishnamurthy · 2026

Read Learning to Reason with Curriculum I: Provable Benefits of Autocurriculum by Nived Rajaraman, Audrey Huang, Miro Dudik, Robert Schapire, Dylan J. Foster, Akshay Krishnamurthy (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2603-18325.txt.

Description

Rajaraman and co-authors prove that autocurriculum helps models learn to reason with less cost. In autocurriculum, a model picks training problems based on its own current performance. For supervised fine-tuning, it needs exponentially fewer reasoning examples than fixed training. Teacher effort lands on prompts where the model now struggles. For reinforcement learning, it splits compute cost from reference model quality, leaving the latter as a near-fixed burn-in cost.

Catalog record

CREATORNived Rajaraman, Audrey Huang, Miro Dudik, Robert Schapire, Dylan J. Foster, Akshay Krishnamurthy
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE120617
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2603.18325

Browse the full library · About the Tower of Babel