RI

#1AutoOR: Scalably Post-training LLMs to Autoformalize Operations Research Problems

MUST-READ

2026-04-18 | X, University of Oxford | 2604.16804

M=8 P=9 I=8

AutoOR trains an 8B LLM to autoformalize linear, mixed-integer, and non-linear optimization problems using a scalable backtranslation data generation pipeline and GRPO with solver-execution rewards. The results are strongly backed by empirical evidence, showing the 8B model outperforming Gemini 2.5 Pro and matching Gemini 3 Pro across 9 benchmarks, including a leap from near 0% to 48.98% on a hard non-linear pump network task. The key insight is the backtranslation pipeline: by instantiating standard OR forms to guarantee ground-truth code and then generating natural language descriptions, the authors bypass the hard problem of verifying code generated from ambiguous text. Furthermore, their curriculum RL strategy—temporarily providing solver syntax as privileged information—successfully breaks the cold-start barrier in hard non-linear domains. This is a must-read for work in OR benchmarking and LLM reasoning evaluation, as the data generation and solver-in-the-loop RL techniques directly address the scalability and verification bottlenecks in symbolic OR modeling.

#2Co-evolving Agent Architectures and Interpretable Reasoning for Automated Optimization

MUST-READ

2026-04-20 | Harbin Institute of Technology, Nanjing University of Information Science and Technology | 2604.17708

M=8 P=8 I=8

Huang et al. propose EvoOR-Agent, a co-evolutionary framework that represents LLM agent workflows as Activity-on-Edge (AOE) networks to simultaneously evolve the agent's architectural topology and its reasoning trajectories for operations research tasks. The results are backed by strong empirical evidence, showing up to 17% improvement over fixed-pipeline OR agents and 15% over general evolutionary agents on complex benchmarks like IndustryOR and BWOR. The key insight is that abstracting agent workflows into an explicit, evolvable AOE graph allows for path-conditioned recombination and structural pruning, enabling the evolutionary search to optimize the problem-solving process (e.g., formulation decomposition, solver routing, debugging loops) rather than just the prompt text or final code. This is highly relevant for our work in LLM evolutionary search and multi-agent optimization, as the AOE graph representation provides a concrete, implementable mechanism for dynamically adapting and evolving agent pipelines.

#3PARM: Pipeline-Adapted Reward Model

MUST-READ

2026-04-20 | The Chinese University of Hong Kong, Hong Kong University of Science and Technology, City University of Hong Kong, Peking University, Tsinghua University, University of California, Los Angeles, Shanghai Jiao Tong University | 2604.18327

M=8 P=8 I=8

Fan et al. introduce a Pipeline-Adapted Reward Model (PARM) that trains stage-specific reward models for LLM optimization pipelines using Direct Preference Optimization on automatically collected execution feedback. The results are backed by strong empirical numbers, demonstrating that a 7B model pipeline can outperform GPT-4o on operations research benchmarks like NL4Opt (0.52 vs 0.15 solving accuracy). The key insight is that intermediate pipeline stages, such as problem formulation, can be effectively scored by training a reward model via DPO where preference pairs are automatically labeled based on whether any downstream execution succeeds. This is highly relevant for our work in multi-agent optimization and LLM evolutionary search, as it provides a concrete, scalable method to train process reward models without human annotation.

#1Learning to Evolve: A Self-Improving Framework for Multi-Agent Systems via Textual Parameter Graph Optimization

MUST-READ

2026-04-22 | Future Living Lab of Alibaba | 2604.20714

M=8 P=8 I=9

He et al. introduce Textual Parameter Graph Optimization (TPGO), a framework that optimizes multi-agent systems by representing their configurations as modular graphs and evolving them using textual gradients. The results are backed by solid empirical evidence, demonstrating a 7-10% absolute success rate improvement on complex benchmarks like MCP-Universe and GAIA over strong baselines (ReAct, MiroFlow). The key insight is the Group Relative Agent Optimization (GRAO) mechanism, which maintains an 'Optimization Experience Memory' of past successful and failed graph edits, clustering and retrieving them to teach the LLM optimizer how to generate better updates over time. This is highly relevant for our work in LLM evolutionary search; adopting the graph-based representation and historical experience retrieval could significantly improve the sample efficiency and stability of our evolutionary algorithms.

#1Hive: A Multi-Agent Infrastructure for Algorithm- and Task-Level Scaling

MUST-READ

2026-04-19 | Peking University | 2604.17353

M=8 P=8 I=8

Hive is an LLM inference infrastructure that optimizes multi-agent and test-time scaling workloads by introducing Logits Cache for redundant sampling paths and Agent-Aware Scheduling for KV cache eviction. The results are backed by solid empirical evidence, demonstrating a 1.11x-1.76x speedup for re-sampling and a 33%-51% reduction in KV cache miss rates on Qwen3-8B. The key insight is that caching intermediate logits (not just KV states) allows the engine to skip expensive forward passes during stochastic resampling of shared prefixes, while evicting KV cache based on an agent's structural contribution outperforms standard LRU. This is highly relevant for scaling LLM evolutionary search and multi-agent optimization, as it provides concrete systems-level techniques to drastically reduce the inference costs associated with branching generation and complex agent coordination.

#2Scepsy: Serving Agentic Workflows Using Aggregate LLM Pipelines

MUST-READ

2026-04-16 | Imperial College London, Independent Researcher | 2604.15186

M=7 P=8 I=8

Scepsy is a serving system that schedules multi-LLM agentic workflows onto GPU clusters by profiling relative LLM execution times to create an Aggregate LLM Pipeline and using a hierarchical heuristic for fractional GPU allocation. The results are strongly backed by empirical numbers on a 16-GPU cluster, showing up to 2.4x higher throughput and 27x lower latency compared to baselines like Kubernetes HPA and Ayo. The key insight is that instead of modeling the highly variable end-to-end latency of dynamic agentic workflows, systems can achieve stable steady-state performance predictions by modeling the aggregate fractional demand each LLM places on the system. This is highly relevant for our research in LLM serving scheduling and GPU resource allocation, providing a strong heuristic baseline and modeling abstraction that our formal OR formulations must benchmark against.