AutoPref: Automatic Discovery of Task-Specific Preference Objectives for Neural Combinatorial Optimization

2026-07-30

AI Summary

This work addresses the limitations of existing neural combinatorial optimization methods, which rely on handcrafted preference objectives that lack generalizability and flexibility. The authors propose AutoPref, a novel framework that automates the discovery of preference objectives for the first time. By leveraging large language models, AutoPref decomposes objectives into pairwise losses and set-aware weighting programs, constructing a unified procedural objective space. A behavior-gated mechanism combined with a staged conditional search strategy enables efficient exploration of this space, supporting both short-horizon training and evaluation. Evaluated across multiple combinatorial optimization problems—including TSP, CVRP, FFSP, and JSSP—AutoPref consistently outperforms strong human-designed baselines across varying problem scales, demonstrating significantly enhanced performance and adaptability.

Abstract

Combinatorial optimization problems (COPs) underpin many real-world decisions, but their exponentially large search spaces make high-quality solutions costly to obtain. Neural combinatorial optimization (NCO) learns fast construction policies, typically with reinforcement learning (RL), while preference-based NCO improves sample efficiency by learning from relative solution quality. However, existing preference objectives combine two distinct design choices in manually specified, one-size-fits-all formulations: what learning signal to extract from each solution pair and how to weight each pair relative to the sampled set. We present AutoPref, the first LLM-guided framework for automated preference-objective discovery in NCO. AutoPref factorizes the objective into a pairwise loss program, which defines the learning signal, and a set-aware weighting program, which determines each pair's relative contribution. Their composition forms a unified programmatic objective space containing existing preference objectives as special cases. To make its search tractable, we introduce a staged conditional search strategy with behavioral gates that filter inadmissible programs before short-horizon training and evaluation. Across TSP, CVRP, FFSP, and JSSP, AutoPref consistently outperforms strong hand-designed baselines across problem scales, demonstrating the benefits and scalability of automated objective discovery for NCO.

Problem

• neural combinatorial optimization
• preference-based learning
• objective design
• automated discovery
• combinatorial optimization

Innovation

• neural combinatorial optimization
• preference-based learning
• automated objective discovery
• programmatic search
• LLM-guided optimization