Toward Ultra-Long-Horizon Agentic Science: Cognitive Accumulation for Machine Learning Engineering
By: Xinyu Zhu, Yuzhu Cai, Zexi Liu, Bingyang Zheng, Cheng Wang, Rui Ye, Jiaao Chen, Hanrui Wang, Wei-Chen Wang, Yuzhi Zhang, Linfeng Zhang, Weinan E, Di Jin, Siheng Chen
Published: 2026-01-15
View on arXiv →#cs.AI
Abstract
This paper introduces ML-Master 2.0, an autonomous agent tackling ultra-long-horizon machine learning engineering. It uses Hierarchical Cognitive Caching to manage context and sustain strategic coherence over long experimental cycles, overcoming limitations of LLMs in real-world research by dynamically distilling transient execution traces into stable knowledge.