Two new research papers introduce advanced frameworks for automated heuristic design in combinatorial optimization problems, leveraging Large Language Models (LLMs) for improved performance. ReVEL utilizes multi-turn reflective LLM guidance with structured performance feedback, organizing heuristics into behavior-aware groups for localized and exploratory refinement. PathWise employs a multi-agent reasoning system with a world model and an entailment graph to plan heuristic generation as a sequential decision process, allowing for state-aware planning and reuse of derivation information. Both approaches demonstrate faster convergence to better heuristics and generalization across different LLM backbones and problem settings. AI
IMPACT These frameworks offer more sophisticated methods for generating effective heuristics in complex optimization tasks, potentially accelerating research and application in fields relying on such problem-solving.
RANK_REASON Two academic papers published on arXiv introduce novel frameworks for automated heuristic design using LLMs.
AI-generated summary · Google Gemini · from 2 sources. How we write summaries →
