arXiv cs.AI· Junbo Jacob Lian, Yujun Sun, Huiling Chen, Chaoyu Zhang, Hanzhang Qin, Chung-Piaw Teo·· 10 天前AI 评分24
ReLoop：面向可靠 LLM 优化的结构化建模与行为验证
ReLoop: Structured Modeling and Behavioral Verification for Reliable LLM-Based Optimization
AI 导读
针对 LLM 生成优化代码易出现"可求解但语义错误"的静默失败问题，ReLoop 提出结构化生成与行为验证双机制：前者将代码生产拆为理解、形式化、合成、验证四阶段推理链，后者通过求解器参数扰动检测残余错误。
正文 · 原文
Abstract:Large language models (LLMs) can translate natural-language problem descriptions into optimization code, but the code is prone to silent failures: it executes and returns a solver-feasible solution while encoding a semantically incorrect formulation. On compositional problems, the resulting feasibility-correctness gap reaches 90 percentage points. We introduce ReLoop, which combines two mechanisms. Structured generation decomposes code production into a four-stage reasoning chain (understand, formalize, synthesize, verify) to reduce formulation errors during generation. Behavioral verification detects the errors that remain by testing whether the formulation responds correctly to solver-based parameter perturbation, a signal that comes from the solver rather than from LLM self-review and requires no ground truth. The two mechanisms address different error structures: structured generation gives the largest gain on compositional problems (+8.5pp accuracy on RetailOpt-190 with Claude Opus 4.6), and behavioral verification gives its largest gain on localized defects (+4.4pp on MAMO-ComplexLP). With diagnostic execution recovery, ReLoop reaches 100% executable code on Claude Opus 4.6, and relative to direct generation it raises or preserves every reported metric of the three chat-tuned foundation models on all three benchmarks. For the narrowly fine-tuned SFT model we test, the chain-of-thought prompt conflicts with its learned output format and lowers its accuracy on MAMO-ComplexLP; we document and analyze this interaction. We release RetailOpt-190, 190 compositional retail optimization scenarios in which several constraints interact.
| Comments: | Code and benchmark: this https URL |
| Subjects: | Software Engineering (cs.SE); Artificial Intelligence (cs.AI); Machine Learning (cs.LG); Optimization and Control (math.OC) |
| Cite as: | arXiv:2602.15983 [cs.SE] |
| (or arXiv:2602.15983v5 [cs.SE] for this version) | |
| https://doi.org/10.48550/arXiv.2602.15983 arXiv-issued DOI via DataCite |
|
| Journal reference: | NeurIPS 2026 |
Submission history
From: Junbo Jacob Lian [view email]
[v1]
Tue, 17 Feb 2026 20:20:33 UTC (192 KB)
[v2]
Wed, 29 Apr 2026 13:39:41 UTC (198 KB)
[v3]
Sun, 16 Aug 2026 14:07:55 UTC (202 KB)
[v4]
Sat, 26 Sep 2026 12:14:35 UTC (205 KB)
[v5]
Tue, 6 Oct 2026 03:07:54 UTC (205 KB)
来源：arXiv cs.AI · arxiv.org