Research4d ago
ReVEL Uses LLMs to Evolve Heuristics
ReVEL groups candidate heuristics into behaviorally coherent clusters, feeding the LLMs concise performance feedback. The framework treats the LLMs as an interactive, multi‑turn reasoner inside an evolutionary algorithm, letting it iteratively refine heuristic rules. This approach reduces brittle one‑shot code synthesis and accelerates the creation of robust NP‑hard problem solvers. Practitioners can adopt ReVEL to automate heuristic engineering.