Changye Li, Zijian Zhang, Muheng Li, Yiduo Pang, Tingjun Wu, Menghan Li, Yitong Liu, Hongkai Guo, Mengchi Chen, Chenyu He, Kaisheng Ma, Wei Xu, Song Han, Sihao Liu, Ligeng Zhu
This paper introduces Humanize MLE Agents (HMA), a workflow-level approach that alternates two agents on a shared workspace to improve long-horizon machine learning engineering. HMA exceeds single-agent limits and achieves state-of-the-art results on 75 Kaggle tasks. It improves any-medal rate over constituent baselines by 6.4 points on average.
The proposed approach has practical significance as it can be applied to various machine learning tasks, improving efficiency and performance. HMA's ability to alternate between agents and reset context enables it to overcome long-horizon deceleration and harness staleness. This can lead to significant advancements in areas like automated machine learning and AI research.
Thank you to arXiv for use of its open access interoperability. arxiv.org. aibrief.ai is not affiliated with or endorsed by arXiv.