Reproduce “ML-AutoResearch: Training Machine Learning Research Agents with Automatically Generated Environments” 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
Ziyang Cai, Amir Saeidi, Harkirat Behl
ML-AutoResearch generates executable machine-learning research tasks, verifies them, and filters teacher-model trajectories into supervised fine-tuning data for agents that edit code, run experiments, and respond to feedback. From 1,000 topics, it produces 555 verified environments and 17,000 examples. Qwen3-8B’s MLGym AUP rises from 0.8915 to 0.9727, a 9.1% relative gain, but results do not establish autonomous scientific discovery.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.