Teacher-Aware Evolution of Heuristic Programs from Learned Optimization Policies

TL;DR

This paper introduces a teacher-aware evolutionary framework that leverages learned optimization policies as behavioral teachers to improve heuristic program discovery for combinatorial optimization tasks.

Contribution

It presents a novel method that uses behavioral feedback from learned policies, rather than direct imitation, to evolve effective heuristics without neural inference at deployment.

Findings

Outperforms performance-driven LLM heuristic evolution baselines.

Requires no neural inference during deployment.

Effective across scheduling, routing, and graph optimization benchmarks.

Abstract

LLM-based automatic heuristic design has shown promise for generating executable heuristics for combinatorial optimization, but existing methods mainly rely on delayed endpoint performance. We propose a \emph{teacher-aware evolutionary framework} that uses independently trained learned optimization policies as behavioral teachers. Instead of deploying or imitating the teacher, our method queries it on states visited by candidate heuristic programs and uses its action preferences as local feedback for evolution. The resulting search discovers static executable heuristics guided by both task performance and teacher-derived behavioral signals. Experiments on scheduling, routing, and graph optimization benchmarks show that our method improves over performance-driven LLM heuristic evolution baselines while requiring no neural inference at deployment. These results suggest that learned optimization policies can be repurposed as behavioral feedback sources for automatic heuristic discovery.