Share your thoughts, 1 month free Claude Pro on usSee more
WorkDL logo mark

HMACE: Heterogeneous Multi-Agent Collaborative Evolution for Combinatorial Optimization

About

Large Language Models have recently emerged as a promising paradigm for automated heuristic design for NP-hard combinatorial optimization problems. Despite this progress, existing LLM-based methods typically rely on monolithic workflows constrained by rigid templates, thereby restricting memory-guided exploration and triggering premature convergence to local optima. To design an autonomous and collaborative architecture, we introduce HMACE, a Heterogeneous Multi-Agent Collaborative Evolution framework that reconceptualizes heuristic search as an organizational design problem. HMACE decomposes each evolutionary generation into an autonomous, role-specialized loop with four coordinated agents: a Proposer for strategy exploration, a Generator for executable heuristic synthesis, an Evaluator for empirical assessment, and a Reflector for archive-backed memory update. By coupling behavior-aware retrieval, lightweight candidate filtering, and fitness-grounded archive updates, HMACE guides the search toward diverse and promising heuristic behaviors while avoiding redundant evaluations. Extensive evaluations on representative COPs, including TSP, Online BPP, MKP, and PFSP, show that HMACE achieves a favorable quality-efficiency trade-off compared to state-of-the-art single-agent and multi-agent baselines. In the matched LLM-driven reference comparison, HMACE achieves the lowest average gaps on TSP and Online BPP (0.464\% and 0.223\%, respectively), while requiring only 0.13M and 0.42M tokens for the two tasks, substantially fewer than the compared baselines.

Yuping Yan, Jirui Han, Fei Ming, Yuanshuai Li, Yaochu Jin• 2026

Related benchmarks

TaskDatasetResultRank
Traveling Salesperson ProblemTSP N=500 Generalization (128 instances)
Optimality Gap0.809
41
Traveling Salesperson ProblemTSP-20 (train)
Objective Value3.84
29
Traveling Salesperson ProblemTSP-50 (train)
Objective Value5.715
29
Online Bin PackingOnline BPP Weibull-5k (test)
Objective Gap0.02
28
Multidimensional Knapsack ProblemMKP
Objective Value46.66
24
Permutation Flow Shop Scheduling ProblemPFSP
Optimality Gap5.73
24
Traveling Salesman ProblemTSP-100 (train)
Objective Value7.767
12
Traveling Salesman ProblemTSP-200 Generalization (test)
Objective Value10.698
12
Traveling Salesman ProblemTSP-1000 Generalization (test)
Objective Value23.409
12
Traveling Salesman ProblemTSP Overall scales 20-1000
Average Gap0.387
11
Showing 10 of 10 rows

Other info

Follow for update