AI that stays on track for days: ML-Master 2.0
AI is great at quick tasks—but stumbles on week-long projects. This paper tackles that ultra-long-horizon gap.
Meet ML-Master 2.0, an autonomous agent for machine learning engineering that stays strategically coherent over days. Its core idea, Hierarchical Cognitive Caching (HCC), treats memory like a multi-level cache and a lab notebook: it condenses messy run-by-run traces into stable know-how and cross-task wisdom, so the agent can execute now while planning for later.
- Short term: keep only what’s needed to act.
- Mid term: distill repeats into playbooks.
- Long term: carry lessons across tasks.
On OpenAI’s MLE-Bench with 24-hour budgets, ML-Master 2.0 sets a new state of the art: a 56.44% medal rate. That’s a concrete step toward agents that can run experiments, learn from sparse feedback, and improve over many cycles—without drowning in context.
Paper: https://arxiv.org/abs/2601.10402v1
Paper: https://arxiv.org/abs/2601.10402v1
Register: https://www.AiFeta.com
AI AutonomousAgents MachineLearning AgenticScience LongHorizon Memory HCC Research arXiv MLEBench MLMaster2
