Automated Heuristic Design

Overview

Automated heuristic design studies how to create efficient rules that guide search and decision-making in problems such as vehicle routing, job scheduling, classical planning, and hardware design. Common approaches use genetic programming, evolutionary search, learned policies, or large language models to generate and refine executable heuristics using solver or simulator feedback. Current work increasingly combines these methods with memory, diverse heuristic portfolios, instance specialization, and transfer across tasks, while evaluating solution quality, runtime, and performance on held-out cases. A central challenge is balancing flexible search with reliable generalization, affordable evaluation, and guarantees such as execution safety or admissibility.

Latest papers

CardsList

92 papers9 papers in the last four weeks, down 36% on the four weeks before. 0.1% of all new papers.

  1. Successor-Generator Planning with LLM-generated Heuristics

    Jan 30, 2025Alexander Tuisov, Yonatan Vernik, Alexander ShleyfmanAutomated Heuristic DesignLLM Planning

  2. Solving Witness Triangle Puzzles Faster with an Automatically Learned Predicate

    Aug 4, 2023Justin Stevens, Vadim Bulitko, David ThueAutomated Heuristic DesignInference-Time Search