Elamur Research⌕
☰
home / papers / 2026 / Reasoning as Gradient: Scaling MLE Agents Beyond Tree Search

Reasoning as Gradient: Scaling MLE Agents Beyond Tree Search

Paper · 2026 · Yifei Zhang, Xu Yang, Xiao Yang et al.

Summary

Most LLM agents for machine learning engineering (MLE) improve solutions through tree or graph search. They reduce rich execution feedback to a scalar score and use it only to decide which candidate to keep. The authors argue that this is a gradient-free approach. It wastes diagnostic information that strong reasoning models could use to work out how a pipeline should change.

They propose Gome, a chain-based agent built around an analogy to optimization:

  • LLM diagnostic reasoning over structured feedback plays the role of the gradient.
  • A shared memory of successful hypotheses acts as momentum.
  • Several parallel traces mimic distributed optimization.

Evaluation uses a closed-world protocol on MLE-Bench, with no external knowledge retrieval. Under these conditions, Gome reaches a state-of-the-art medal rate on modest hardware.

Experiments across 10 LLMs show a crossover. Search-based agents remain better with weaker models, but Gome's lead grows as reasoning ability increases. The authors conclude that improving the quality of the update signal may matter more than designing better search strategies.

Key points

  • On MLE-Bench (75 Kaggle competitions), Gome reaches a 35.1% any-medal rate when limited to 12 hours of running time on one V100 GPU.
  • Each iteration has four stages: execution, hierarchical validation, memory update and reasoning. Validation checks output format, then data leakage and overfitting, then overall quality.
  • Baselines include AIDE, ML-Master and AIRA, plus Gome-MCTS, an MCTS variant built on the same codebase.
  • Gome trails tree search on efficiency-tier models such as GPT-4o. On frontier models, its advantage over search-based baselines widens to +5.8% with o3 and +7.1% with GPT-5.
  • The code and GPT-5 execution traces are available in the microsoft/RD-Agent GitHub repository. Stated limitations include reliance on strong (and costlier) models and no formal convergence guarantees.

Our summary of the paper, in our own words — not its text. The work belongs to its authors; read it at the links above.

 Fields, topics and concepts

📚 Shelves 5The field of science this work belongs to (OpenAlex fields).Computer Science · 18,614 worksArtificial Intelligence · 17,965 worksArts and Humanities · 17,450 worksLanguage and Linguistics · 17,296 worksInformation Systems · 1,730 works

🧭 Topics 1Research subjects and languages the work is about.Reasoning · 1,685 works

Cite

Cite as Yifei Zhang, Xu Yang, Xiao Yang, Bowen Xian, Qizheng Li, Shikai Fang, Jingyuan Li, Jian Wang, Minrui Xu, Yuge Zhang, Weiqing Liu and Jiang Bian. 2026. Reasoning as Gradient: Scaling MLE Agents Beyond Tree Search. Findings of the Association for Computational Linguistics: ACL 2026.

The paper itself, not this page. BibTeX from the library's record of the work. DOI 10.18653/v1/2026.findings-acl.438

@inproceedings{zhang2026reasoning,
  title = {{Reasoning as Gradient: Scaling MLE Agents Beyond Tree Search}},
  author = {Yifei Zhang, Xu Yang, Xiao Yang, Bowen Xian, Qizheng Li, Shikai Fang, Jingyuan Li, Jian Wang, Minrui Xu, Yuge Zhang, Weiqing Liu and Jiang Bian},
  booktitle = {Findings of the Association for Computational Linguistics: ACL 2026},
  year = {2026},
  doi = {10.18653/v1/2026.findings-acl.438},
  url = {https://doi.org/10.18653/v1/2026.findings-acl.438}
}

References

35 references, 13 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. Artificial Analysis Team. 2025. Artificial Analysis: Independent Benchmarks and Performance Landscape of AI Models. Accessed: 2025-12-28. find
  2. Saeed Asadi, Sonia Gharibzadeh, Hajar Kazemi Naeini, Masoud Reihanifar, Morteza Rahimi, Shiva Zangeneh, Aseel Smerat, and Lazim Abdullah. 2024. Comparative analysis of gradient-based optimization techniques using multidimensional surface 3d visualizations and initial point sensitivity. arXiv preprin… — [arXiv:2409.04470](https://arxiv.org/abs/2409.04470) find
  3. Léon Bottou, Frank E Curtis, and Jorge Nocedal. 2018. Optimization methods for large-scale machine learning. SIAM review, 60(2):223–311. find
  4. Jun Shern Chan, Neil Chowdhury, Oliver Jaffe, James Aung, Dane Sherburn, Evan Mays, Giulio Starace, Kevin Liu, Leon Maksin, Tejal Patwardhan, and 1 others. 2024. Mle-bench: Evaluating machine learning agents on machine learning engineering. arXiv preprint arXiv:2410.07095.
  5. Ching-An Cheng, Allen Nie, and Adith Swaminathan. 2024. Trace is the next autodiff: Generative optimization with rich feedback, execution traces, and llms. Advances in Neural Information Processing Systems, 37:71596–71642. find
  6. Anthony Cui, Pranav Nandyalam, Andrew Rufail, Ethan Cheung, Aiden Lei, Kevin Zhu, and Sean O’Brien. 2024. Introducing mapo: Momentum-aided gradient descent prompt optimization. arXiv preprint arXiv:2410.19499. — [arXiv:2410.19499](https://arxiv.org/abs/2410.19499) find
  7. Shangheng Du, Xiangchao Yan, Dengyang Jiang, Jiakang Yuan, Yusong Hu, Xin Li, Liang He, Bo Zhang, and LEI BAI. Internagent-mle: Navigating fine-grained optimization for coding agent. find
  8. Timur Garipov, Pavel Izmailov, Dmitrii Podoprikhin, Dmitry P Vetrov, and Andrew G Wilson. 2018. Loss surfaces, mode connectivity, and fast ensembling of dnns. Advances in neural information processing systems, 31. find
  9. Paul Gauthier. 2025. Aider LLM Leaderboards: Code Editing and Refactoring Benchmarks. https://aider.chat/docs/leaderboards/. Accessed: 2025-12-28. ↗
  10. Daya Guo, Dejian Yang, Haowei Zhang, Junxiao Song, Ruoyu Zhang, Runxin Xu, Qihao Zhu, Shirong Ma, Peiyi Wang, Xiao Bi, and 1 others. 2025. Deepseek-r1: Incentivizing reasoning capability in llms via reinforcement learning. arXiv preprint arXiv:2501.12948.
  11. Qingyan Guo, Rui Wang, Junliang Guo, Bei Li, Kaitao Song, Xu Tan, Guoqing Liu, Jiang Bian, and Yujiu Yang. 2023. Evoprompt: Connecting llms with evolutionary algorithms yields powerful prompt optimizers. arXiv e-prints, pages arXiv–2309.
  12. Qian Huang, Jian Vora, Percy Liang, and Jure Leskovec. 2023. Mlagentbench: Evaluating language agents on machine learning experimentation. arXiv preprint arXiv:2310.03302.
  13. Akira Isihara. 2013. Statistical physics. Academic Press. find
  14. Aaron Jaech, Adam Kalai, Adam Lerer, Adam Richardson, Ahmed El-Kishky, Aiden Low, Alec Helyar, Aleksander Madry, Alex Beutel, Alex Carney, and 1 others. 2024. Openai o1 system card. arXiv preprint arXiv:2412.16720.
  15. Zhengyao Jiang, Dominik Schmidt, Dhruv Srikanth, Dixing Xu, Ian Kaplan, Deniss Jacenko, and Yuxiang Wu. 2025. Aide: Ai-driven exploration in the space of code. arXiv preprint arXiv:2502.13138. find
  16. Carlos E Jimenez, John Yang, Alexander Wettig, Shunyu Yao, Kexin Pei, Ofir Press, and Karthik Narasimhan. 2023. Swe-bench: Can language models resolve real-world github issues? arXiv preprint arXiv:2310.06770.
  17. Stepan Kulibaba, Artem Dzhalilov, Roman Pakhomov, Oleg Svidchenko, Alexander Gasnikov, and Aleksei Shpilman. 2025. Kompeteai: Accelerated autonomous multi-agent system for end-to-end pipeline generation for machine learning problems. arXiv preprint arXiv:2508.10177. — [arXiv:2508.10177](https://arxiv.org/abs/2508.10177) find
  18. Yurii Nesterov and 1 others. 2018. Lectures on convex optimization, volume 137. Springer. find
  19. Fan Nie, Junlin Wang, Harper Hua, Federico Bianchi, Yongchan Kwon, Zhenting Qi, Owen Queen, Shang Zhu, and James Zou. 2026. Dsgym: A holistic framework for evaluating and training data science agents. arXiv preprint arXiv:2601.16344. — [arXiv:2601.16344](https://arxiv.org/abs/2601.16344) find
  20. OpenAI. 2025. Introducing openai o3 and o4-mini. Accessed: 2025-12-22. find
  21. Reid Pryzant, Dan Iter, Jerry Li, Yin Tat Lee, Chenguang Zhu, and Michael Zeng. 2023. Automatic prompt optimization with" gradient descent" and beam search. arXiv preprint arXiv:2305.03495.
  22. Noah Shinn, Federico Cassano, Ashwin Gopinath, Karthik Narasimhan, and Shunyu Yao. 2023. Reflexion: Language agents with verbal reinforcement learning. Advances in Neural Information Processing Systems, 36:8634–8652.
  23. Aaditya Singh, Adam Fry, Adam Perelman, Adam Tart, Adi Ganesh, Ahmed El-Kishky, Aidan McLaughlin, Aiden Low, AJ Ostrow, Akhila Ananthram, and 1 others. 2025. Openai gpt-5 system card. arXiv preprint arXiv:2601.03267.
  24. Jiankai Sun, Chuanyang Zheng, Enze Xie, Zhengying Liu, Ruihang Chu, Jianing Qiu, Jiaqi Xu, Mingyu Ding, Hongyang Li, Mengzhe Geng, and 1 others. 2025. A survey of reasoning with foundation models: Concepts, methodologies, and outlook. ACM Computing Surveys, 57(11):1–43. find
  25. Annan Li, Chufan Wu, Zengle Ge, Yee Hin Chong, Zhinan Hou, Lizhe Cao, Cheng Ju, Jianmin Wu, Huaiming Li, Haobo Zhang, and 1 others. 2025. The fm agent. arXiv preprint arXiv:2510.26144. find
  26. Edan Toledo, Karen Hambardzumyan, Martin Josifoski, Rishi Hazra, Nicolas Baldwin, Alexis Audran-Reiss, Michael Kuchnik, Despoina Magka, Minqi Jiang, Alisia Maria Lupidi, and 1 others. 2025. Ai research agents for machine learning: Search, exploration, and generalization in mle-bench. arXiv preprint… — [arXiv:2507.02554](https://arxiv.org/abs/2507.02554) find
  27. Zexi Liu, Yuzhu Cai, Xinyu Zhu, Yujie Zheng, Runkun Chen, Ying Wen, Yanfeng Wang, Siheng Chen, and 1 others. 2025. Ml-master: Towards ai-for-ai via integration of exploration and reasoning. arXiv preprint arXiv:2506.16499. — [arXiv:2506.16499](https://arxiv.org/abs/2506.16499) find
  28. Xingyao Wang, Boxuan Li, Yufan Song, Frank F Xu, Xiangru Tang, Mingchen Zhuge, Jiayi Pan, Yueqi Song, Bowen Li, Jaskirat Singh, and 1 others. 2024. Openhands: An open platform for ai software developers as generalist agents. arXiv preprint arXiv:2407.16741.
  29. Aman Madaan, Niket Tandon, Prakhar Gupta, Skyler Hallinan, Luyu Gao, Sarah Wiegreffe, Uri Alon, Nouha Dziri, Shrimai Prabhumoye, Yiming Yang, and 1 others. 2023. Self-refine: Iterative refinement with self-feedback. Advances in Neural Information Processing Systems, 36:46534–46594. find
  30. Fengli Xu, Qianyue Hao, Zefang Zong, Jingwei Wang, Yunke Zhang, Jingyi Wang, Xiaochong Lan, Jiahui Gong, Tianjian Ouyang, Fanjin Meng, and 1 others. 2025. Towards large reasoning models: A survey of reinforced reasoning with large language models. arXiv preprint arXiv:2501.09686.
  31. Niru Maheswaranathan, Luke Metz, George Tucker, Dami Choi, and Jascha Sohl-Dickstein. 2019. Guided evolutionary strategies: Augmenting random search with surrogate gradients. In International Conference on Machine Learning, pages 4264–4273. PMLR. find
  32. Chengrun Yang, Xuezhi Wang, Yifeng Lu, Hanxiao Liu, Quoc V Le, Denny Zhou, and Xinyun Chen. 2023. Large language models as optimizers. In The Twelfth International Conference on Learning Representations.
  33. Jaehyun Nam, Jinsung Yoon, Jiefeng Chen, Jinwoo Shin, Sercan Ö Arık, and Tomas Pfister. 2025. Mle-star: Machine learning engineering agent via search and targeted refinement. arXiv preprint arXiv:2506.15692. — [arXiv:2506.15692](https://arxiv.org/abs/2506.15692) find
  34. Mert Yuksekgonul, Federico Bianchi, Joseph Boen, Sheng Liu, Zhi Huang, Carlos Guestrin, and James Zou. 2024. Textgrad: Automatic" differentiation" via text. arXiv preprint arXiv:2406.07496.
  35. Wenlin Zhang, Xiangyang Li, Kuicai Dong, Yichao Wang, Pengyue Jia, Xiaopeng Li, Yingyi Zhang, Derong Xu, Zhaocheng Du, Huifeng Guo, and 1 others. 2025. Process vs. outcome reward: Which is better for agentic rag reinforcement learning. arXiv preprint arXiv:2505.14069. — [arXiv:2505.14069](https://arxiv.org/abs/2505.14069) find

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/zhang-2026-reasoning-as-gradient-scaling-mle/