This work shows empirically that solver-integrated reasoning, exact combinatorial algorithm, and heuristic search exhibit complementary strengths across different problem structures and scales, and introduces Strategy-Diverse Reinforcement Learning (SDRL), which trains LLMs as adaptive optimization meta-solvers.
Abstract
Scaling LLM-based optimization from textbook-scale instances to real-world, industrial tasks remains a critical open challenge. Existing approaches are predominantly evaluated on small, self-contained textual problems and often commit to a solver-integrated paradigm, limiting their ability to handle the scale and structural diversity of practical optimization workloads. In this work, we propose a practical framework for training open-source LLMs to tackle real-world, industrial-scale optimization. We first show empirically that solver-integrated reasoning, exact combinatorial algorithm, and heuristic search exhibit complementary strengths across different problem structures and scales. Motivated by this, we introduce Strategy-Diverse Reinforcement Learning (SDRL), which trains LLMs as adaptive optimization meta-solvers. SDRL leverages this complementarity through a correctness-gated hierarchical diversity reward that promotes robust exploration across varying strategies and within each strategy, effectively preventing premature strategy collapse. We further introduce a mixed-format training scheme that jointly supports both self-contained textual problems and file-grounded instances. Across comprehensive evaluations, our framework outperforms existing fine-tuned methods and frontier models including DeepSeek-V4-Pro and GPT-5.5, both on average across benchmarks and on industrial-scale optimization tasks.
A challenger-solver asymmetric self-play approach, where an LLM challenger generates progressively harder instances and the solver verifies feasibility and hardness, so no seed or training MILP instances are required.
Jitin Singla, Parikshit Pareek, P. Jawanpuria et al.· 0 citations
This work investigates a data-driven selection of the L-BFGS memory parameter (mem) for large-scale unconstrained optimization and proposes a two-stage hybrid machine-learning framework that predicts intermediate solver quantities and then forecasts runtime for each candidate memory value.
Yu-Chen Gao· Journal of Computing, Data,...· 0 citations
A solver-informed diagnosis mechanism that exploits fine-grained solver statistics as verbal gradients for targeted refinement and a structured memory abstracts prior experience into reusable modeling strategies, avoiding redundant exploration while enabling zero-shot transfer to unseen problems and bootstrapping small...
Haofeng Yuan, Jianing Peng, Jieyi Bi et al.· 0 citations
This paper formalizes test suite generation as a Markov Decision Process (MDP) and demonstrates that its objective exhibits monotone submodularity, which enables an effective relaxation of this NP-hard global optimization into a tractable step-wise greedy procedure.
Guo-Qing Wang, Cheng-Ran Yang, Xiao-Xuan Zhou et al.· Proceedings of the ACM on So...· 0 citations
This paper introduces AESAT (Auto-Evolving SAT solving), a novel neuro-symbolic framework designed to automatically evolve and optimize the heuristic functions of SAT solvers, and represents the first time an LLM-enhanced solver has dominated the world's premier SAT competition.
Mao Luo, Hang Ding, Chumin Li et al.· Proceedings of the Thirty-Fi...· 0 citations
Results indicate that with zero curated data, OPT-Zero matches state-of-the-art data-dependent methods while exhibiting substantially stronger generalizability, establishing self-play training as a highly scalable paradigm for advancing LLM reasoning in modeling and solving optimization problems.
By studying the computational foundations of language learning, Associate Professor Jacob Andreas builds intelligent systems that can communicate effectively with humans.
MIT News · Artificial Intelligence· news.mit.eduAug 27, 2026
A new machine-learning framework aims to improve the success rate of computational protein design while moving away from results that reproduce sequences found in nature.