Elamur Research⌕
☰
home / papers / 2026 / Escaping the Sisyphus Dilemma: Experience Replay for Robust Text-to-Optimization Modeling

Escaping the Sisyphus Dilemma: Experience Replay for Robust Text-to-Optimization Modeling

Paper · 2026 · Wantong Xie, Yinghao Chen, Yi-Xiang Hu et al.

Summary

LLMs can turn natural-language problem descriptions into executable optimization models, such as MILP formulations for solvers like Gurobi. They usually repair errors through repeated exchanges with the solver, but they keep nothing between sessions. As a result, the same mistakes come back on structurally similar problems, which the authors call the Sisyphus Dilemma. Standard retrieval-augmented methods only supply finished problem-model examples. These show what a correct answer looks like, not how to repair a broken model.

The paper proposes EOM (Experience for Optimization Modeling), an experience-replay framework. After each session, an LLM distiller condenses the interaction log into a Causal Correction Mapping. This links a symptom to an insight:

  • Symptom: problem semantics, a linearized AST of the faulty code, and the solver traceback.
  • Insight: a diagnosis and targeted patch for successful sessions, or traps and warnings for failed ones.

A structure-aware retriever then matches new errors against this growing pool. Across seven benchmarks, EOM improves accuracy on complex tasks while lowering token usage and dialogue turns. This suggests that persistent error memory makes LLM-based OR modeling more reliable and cheaper.

Key points

  • Reported gains: +8.45% modeling accuracy on complex benchmarks, 28.65% fewer tokens, and 25.82% fewer interaction turns, evaluated on seven benchmarks.
  • Two nested loops make up the design: the outer one, called the Execution Loop, uses solver feedback to refine models, while the inner Experience Loop builds up distilled knowledge over time.
  • Each experience unit pairs a hybrid symptom vector (semantic embedding, AST signature that ignores variable names, solver traceback) with a corrective insight.
  • Failed sessions are kept as negative knowledge ("Traps" and "Warnings") to narrow future search, alongside positive Diagnosis/Prescription pairs.
  • Limitations: EOM depends on explicit solver errors, so it cannot catch silent semantic mistakes, and it is vulnerable to noisy or hallucinated distilled experiences. Planned future work covers filtering low-quality experiences and extending to nonlinear optimization.

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 4The 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 works

Cite

Cite as Wantong Xie, Yinghao Chen, Yi-Xiang Hu, Feng Wu, Jieyang Xu, Sijia Zhang and Xiangyang Li. 2026. Escaping the Sisyphus Dilemma: Experience Replay for Robust Text-to-Optimization Modeling. 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.690

@inproceedings{xie2026escaping,
  title = {{Escaping the Sisyphus Dilemma: Experience Replay for Robust Text-to-Optimization Modeling}},
  author = {Wantong Xie, Yinghao Chen, Yi-Xiang Hu, Feng Wu, Jieyang Xu, Sijia Zhang and Xiangyang Li},
  booktitle = {Findings of the Association for Computational Linguistics: ACL 2026},
  year = {2026},
  doi = {10.18653/v1/2026.findings-acl.690},
  url = {https://doi.org/10.18653/v1/2026.findings-acl.690}
}

References

28 references, 6 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. Ali AhmadiTeshnizi, Wenzhi Gao, Herman Brunborg, Shayan Talaei, Connor Lawless, and Madeleine Udell. 2024a. Optimus-0.3: Using large language models to model and solve optimization problems at scale. arXiv preprint arXiv:2407.19633. — [arXiv:2407.19633](https://arxiv.org/abs/2407.19633) find
  2. Ali AhmadiTeshnizi, Wenzhi Gao, and Madeleine Udell. 2024b. Optimus: Scalable optimization modeling with (mi) lp solvers and large language models. In Forty-first International Conference on Machine Learning. find
  3. Akari Asai, Zeqiu Wu, Yizhong Wang, Avirup Sil, and Hannaneh Hajishirzi. 2024. Self-rag: Learning to retrieve, generate, and critique through self-reflection. In The Twelfth International Conference on Learning Representations.
  4. Nicolás Astorga, Tennison Liu, Yuanzhang Xiao, and Mihaela van der Schaar. 2025. Autoformulation of mathematical optimization models using llms. In Forty-second International Conference on Machine Learning. find
  5. Albert Camus. 2013. The myth of Sisyphus. Penguin UK. find
  6. Xinyun Chen, Maxwell Lin, Nathanael Schärli, and Denny Zhou. 2023. Teaching large language models to self-debug. In The Twelfth International Conference on Learning Representations. find
  7. DeepSeek-AI. 2025. Deepseek-v3.2: Pushing the frontier of open large language models.
  8. Gurobi Optimization, LLC. 2025. Gurobi Optimizer Reference Manual. find
  9. Chenyu Huang, Zhengyang Tang, Shixi Hu, Ruoqing Jiang, Xin Zheng, Dongdong Ge, Benyou Wang, and Zizhuo Wang. 2025. Orlm: A customizable framework in training large models for automated optimization modeling. Operations Research. find
  10. Jie Huang, Xinyun Chen, Swaroop Mishra, Huaixiu Steven Zheng, Adams Wei Yu, Xinying Song, and Denny Zhou. 2024a. Large language models cannot self-correct reasoning yet. In The Twelfth International Conference on Learning Representations.
  11. Xuhan Huang, Qingning Shen, Yan Hu, Anningzhe Gao, and Benyou Wang. 2024b. Mamo: a mathematical modeling benchmark with solvers. arXiv preprint arXiv:2405.13144. find
  12. Caigao Jiang, Xiang Shu, Hong Qian, Xingyu Lu, JUN ZHOU, Aimin Zhou, and Yang Yu. 2025. Llmopt: Learning to define and solve general optimization problems from scratch. In The Thirteenth International Conference on Learning Representations. find
  13. Hunter Lightman, Vineet Kosaraju, Yuri Burda, Harrison Edwards, Bowen Baker, Teddy Lee, Jan Leike, John Schulman, Ilya Sutskever, and Karl Cobbe. 2023. Let’s verify step by step. In The Twelfth International Conference on Learning Representations.
  14. Haoyang Liu, Jie Wang, Yuyang Cai, Xiongwei Han, Yufei Kuang, and Jianye HAO. 2025a. Optitree: Hierarchical thoughts generation with tree search for llm optimization modeling. In The Thirty-ninth Annual Conference on Neural Information Processing Systems. find
  15. Yitao Liu, Chenglei Si, Karthik R Narasimhan, and Shunyu Yao. 2025b. Contextual experience replay for self-improvement of language agents. In Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), pages 14179–14198. find
  16. Hongliang Lu, Zhonglin Xie, Yaoyu Wu, Can Ren, Yuxuan Chen, and Zaiwen Wen. 2025. Optmath: A scalable bidirectional data synthesis framework for optimization modeling. arXiv preprint arXiv:2502.11102. learning. Advances in Neural Information Processing Systems, 36:8634–8652. — [arXiv:2502.11102](https://arxiv.org/abs/2502.11102) find
  17. Segev Wasserkrug, Leonard Boussioux, Dick Den Hertog, Farzaneh Mirzazadeh, Ş Ilker Birbil, Jannis Kurtz, and Donato Maragno. 2025. Enhancing decision making through the integration of large language models and operations research optimization. In Proceedings of the AAAI Conference on Artificial Int… find
  18. Jason Wei, Xuezhi Wang, Dale Schuurmans, Maarten Bosma, Fei Xia, Ed Chi, Quoc V Le, Denny Zhou, and 1 others. 2022. Chain-of-thought prompting elicits reasoning in large language models. Advances in neural information processing systems, 35:24824–24837.
  19. Ziyang Xiao, Dongxiang Zhang, Yangjun Wu, Lilin Xu, Yuan Jessica Wang, Xiongwei Han, Xiaojin Fu, Tao Zhong, Jia Zeng, Mingli Song, and 1 others. 2023. Chain-of-experts: When llms meet complex operations research problems. In The twelfth international conference on learning representations. find
  20. Wantong Xie, Yi-Xiang Hu, Jieyang Xu, Feng Wu, and Xiangyang Li. 2025. Murka: Multi-reward reinforcement learning with knowledge alignment for optimization tasks. In The Thirty-ninth Annual Conference on Neural Information Processing Systems. find
  21. 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
  22. Zhicheng Yang, Yiwei Wang, Yinya Huang, Zhijiang Guo, Wei Shi, Xiongwei Han, Liang Feng, Linqi Song, Xiaodan Liang, and Jing Tang. 2024. Optibench meets resocratic: Measure and improve llms for optimization modeling. In The Thirteenth International Conference on Learning Representations. find
  23. Stephen Maher, Matthias Miltenberger, João Pedro Pedroso, Daniel Rehfeldt, Robert Schwarz, and Felipe Serrano. 2016. PySCIPOpt: Mathematical programming in python with the SCIP optimization suite. In Mathematical Software – ICMS 2016, pages 301– 307. Springer International Publishing. find
  24. Andrew Zhao, Daniel Huang, Quentin Xu, Matthieu Lin, Yong-Jin Liu, and Gao Huang. 2024. Expel: Llm agents are experiential learners. In Proceedings of the AAAI Conference on Artificial Intelligence, volume 38, pages 19632–19642. find
  25. Xiaotian Pan, Junhao Fang, Feng Wu, Sijia Zhang, Yi-Xiang Hu, Shaoang Li, and Xiang-Yang Li. 2025. Guiding large language models in modeling optimization problems via question partitioning. In Proceedings of the Thirty-Fourth International Joint Conference on Artificial Intelligence, pages 2657– 266… find
  26. Lily Zhong, Zilong Wang, and Jingbo Shang. 2024. Ldb: A large language model debugger via verifying runtime execution step-by-step. arXiv preprint arXiv:2402.16906.
  27. Rindranirina Ramamonjison, Timothy Yu, Raymond Li, Haley Li, Giuseppe Carenini, Bissan Ghaddar, Shiqi He, Mahdi Mostajabdaveh, Amin Banitalebi-Dehkordi, Zirui Zhou, and 1 others. 2023. Nl4opt competition: Formulating optimization problems based on their natural language descriptions. In NeurIPS 2022… find
  28. Noah Shinn, Federico Cassano, Ashwin Gopinath, Karthik Narasimhan, and Shunyu Yao. 2023. Reflexion: Language agents with verbal reinforcement find

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/xie-2026-escaping-sisyphus-dilemma-experience-replay/