VALG: An open-source agentic system for automated ML theory research

YiMaTweets · x · 2026-08-17

VALG is an open-source agentic system designed for automated ML theory research, converting the iterative rhythm of theoretical work into a structured workflow. A key feature is joint problem solving and problem reformulation, iterating between adjusting assumptions and deriving proofs.

Tested on 5 COLT 2026 open problems, VALG produced 22 theorem candidates across 9 subproblems. Two were fully solved, and seven were partially resolved via relaxations. The team is also building a benchmark for ML theory automation, with initial results suggesting around 80% of arXiv 2026 theory papers can be handled end-to-end.

Original post →

More from coding & agent

coding & agent channel →