RI

#1Large Language Model for Operations Research Formulation Selection in Multi-Warehouse Inventory Allocation

2026-07-28 | JD.com |2607.25956

Xu et al. propose an LLM-based selector trained via SFT, IPO and GRPO to route multi-warehouse inventory allocation instances to the most suitable mixed-integer programming (MIP) formulation. Backed by real-world data from JD.com, the GRPO-trained selector improves top-1 expert selection accuracy by 29 percentage points over an SFT+IPO baseline, translating to a 12.5% allocation accuracy gain over the incumbent system. The key insight is the offline GRPO metadata construction: pre-computing and caching solver evaluations (scores, rankings, best expert) for historical instances to serve as fast, reliable reward signals during iterative RL post-training, which avoids expensive solver calls in the RL loop.

#2Search Hardness-Aware LLM-Based Problem Formulation for Expensive Simulation-Driven Design

2026-07-23 | Xidian University, Victoria University of Wellington |2607.21220

This paper proposes an LLM-based evolutionary framework to automatically generate and refine optimization problem formulations (objectives and constraints) for expensive simulation-driven design. The authors empirically demonstrate on a hydrology task and five antenna design benchmarks that their evolved formulations require significantly fewer expensive simulator evaluations to reach feasible designs compared to expert-designed or zero-shot LLM formulations. The key insight is using initial random simulation data to identify 'hard but promising' (rare and non-dominated) states and then using a candidate formulation's ability to prioritize these anchor states as the fitness signal for the evolutionary search. This is highly relevant for research in LLM evolutionary search and automated OR modeling, as it provides a concrete method for constructing proxy rewards that improve sample efficiency when evaluating generated code is computationally expensive.

#1SpecAHD: Localize to Specialize for Automated Heuristic Design in Large-Scale Routing Problems

2026-07-26 | Guangdong University of Technology |2607.23676

SpecAHD introduces a bilevel LLM evolutionary search framework that co-evolves an upper-level program to expose local repair regions and a lower-level repertoire of specialized constructive heuristics to repair them. The results are rigorously backed by extensive empirical evidence, showing cost reductions of up to 57.7% against state-of-the-art AHD baselines (like EoH and ReEvo) on large-scale routing benchmarks including CVRP and VRPTW up to 10,000 nodes. The key insight is the 'Repair Response Archive'—a memory mechanism that logs heuristic performance on specific local structures during training, allowing a k-NN router to dynamically select the best specialized heuristic for new regions at inference time.

#2Teaching LLMs to Self-Evolve: Cultivating Core Meta-Skills with Reinforcement Learning

2026-07-24 | University of Illinois Urbana-Champaign |2607.21971

Wu et al. introduce METAEVOLVE, a framework that uses reinforcement learning (GRPO) to explicitly train LLMs to perform iterative self-evolution and refinement. The results are rigorously backed by execution-based metrics, showing a 24% absolute improvement over AlphaEvolve on out-of-distribution coding tasks and a 46% relative speedup on open-ended algorithm optimization benchmarks. The key insight is that instead of relying on zero-shot prompting in an evolutionary loop, researchers can synthesize evolution trajectories (current code, fitness, history) from static datasets and use execution feedback as a verifiable reward to fine-tune the mutator LLM.

#3Budget-Aware LLM Discovery via Cost-Calibrated Frontier Utility

2026-07-29 | Mohamed bin Zayed University of Artificial Intelligence (MBZUAI), City University of Hong Kong, International Digital Economy Academy (IDEA) |2607.26828

This paper introduces CostAda, an adaptive controller for LLM evolutionary search that allocates compute across search frontiers based on cost-calibrated utility (progress divided by realized token cost) and remaining budget. The results are rigorously backed by empirical evaluations across 8 benchmarks using GLM-5 and GPT-5.4, demonstrating that CostAda reaches the full-budget quality of state-of-the-art baselines (AdaEvolve, EvoX) using only half the token budget. The key insight is that progress-only credit is fundamentally flawed when search actions (e.g., short refinements vs. long-context guidance) have vastly different token costs; dividing fitness gains by the log of realized cost and shifting weight from local to global progress as the budget drains drastically improves search efficiency.

#4EvoPINN: Agentic Discovery of Executable Algorithms for Physics-Informed Neural Networks

2026-07-29 | Institute of Automation, Chinese Academy of Sciences, University of Chinese Academy of Sciences |2607.26490

EvoPINN is an LLM-driven evolutionary framework that autonomously discovers novel neural architectures and training programs for Physics-Informed Neural Networks (PINNs). Backed by rigorous empirical results, the framework evaluated only 56 candidates to discover algorithms that outperformed expert-designed baselines by up to 35% on complex PDE benchmarks. The key insight is its decoupled, module-wise evolution strategy: it uses a UCB scheduler to dynamically allocate LLM mutation budget between different algorithm components, combined with AST-based structural verification to filter out cosmetic LLM edits before triggering expensive evaluations. This is highly relevant for LLM evolutionary search research, as the telemetry-guided generation and modular credit assignment directly address the sample efficiency and sparse reward challenges inherent in automated heuristic discovery.

#5VPR-Evolve: Multi-Agent-Driven Algorithm Evolution for FPGA Place and Route

2026-07-27 | Arizona State University |2607.24998

Wu et al. introduce VPR-Evolve, a multi-agent LLM framework that specializes FPGA place-and-route algorithms by directly evolving the tool's source code for specific hardware designs. The results are backed by rigorous empirical evaluation on VTR-9 benchmarks, showing up to a 9.8% reduction in critical-path delay and outperforming Bayesian hyperparameter tuning with significantly fewer evaluations. The key insight is the framework's architecture, which uses staged evolution to optimize different algorithmic components sequentially and employs an 'Inspiration Collector' agent that injects new literature-based ideas specifically when the search plateaus.

#6A Control System, a Dataset and a Recipe for Making Frozen LLM Agents Learn a Domain

2026-07-28 |2607.25415

This paper formulates LLM agent harness optimization (prompts, tools, memory, planning) as a discrete control problem and evaluates online RL (bandits, REINFORCE) against a static DSPy-optimized baseline. Backed by rigorous empirical results across multiple domains and models, the authors find that online RL consistently underperforms the static DSPy baseline within practical episode budgets (e.g., 62% vs 96% success on tool-use). The key insight is that starting online controllers from a uniform prior over a discrete harness space is severely sample-inefficient; dynamic harness optimization must be bootstrapped from a metric-guided static optimizer to be viable.

#7NeurGO: Learning to Generate Elite Candidates for Meta-Black-Box Expensive Optimization

2026-07-26 | China University of Geosciences |2607.23408

NeurGO is a generative meta-black-box optimization framework that uses a Transformer encoder to capture population search trends and an MLP decoder to directly synthesize a small set of elite candidates for expensive optimization tasks. The method is backed by strong empirical results on CEC 2008 and COCO BBOB benchmarks (D=10), demonstrating faster convergence and better final performance than Bayesian Optimization and other MetaBBO baselines under a strict 300-evaluation budget. The key insight is that instead of generating raw candidate coordinates, the decoder outputs contribution weights used to recombine the existing population's representations, ensuring generated candidates remain anchored to known promising regions and avoiding out-of-distribution proposals. This architectural design is highly relevant for continuous latent-space optimization and could be directly adapted to improve sample efficiency when optimizing continuous representations of algorithms or heuristics.

#1TileSight: A First-Principles Tile-Centric Analytical GPU Performance Model from Cores to Clusters

2026-07-24 | Imperial College London, Peking University, Microsoft Research, University of Edinburgh, Shanghai Jiao Tong University, Tile-AI |2607.22432

TileSight is a first-principles, tile-centric analytical performance model that predicts GPU kernel and end-to-end LLM serving latency by simulating intra-tile resource usage, inter-tile cache reuse and cross-device communication. The results are rigorously backed by hardware measurements, achieving 12.35% MAPE on single-GPU kernels and 13.52% wMAPE on end-to-end vLLM serving across diverse architectures (A100 to B6000), outperforming learned predictors. The key insight is that lifting performance modeling to the tile abstraction (rather than thread or cache-line level) enables fast, schedule-sensitive and deterministic latency predictions without requiring per-architecture ML training. This is highly relevant for research in LLM serving scheduling and GPU resource allocation, as TileSight can serve as a highly accurate, white-box cost estimator for optimization formulations.