매일 아침, 어제의 AI를 한 통으로 정리해 보내드립니다메일로 받아보기

METAL LAB

Evolving Parallel Algorithm Portfolios via Potential-Aware Instance Generation with LLMs

arXiv:2608.068082026-08-10

arXiv:2608.06808v1 Announce Type: new Abstract: The Automatic Construction of Portfolios via Large Language Models (LLM-ACP) suffers from poor generalization in practical few-shot scenarios when solving complex combinatorial optimization problems. Instance and algorithm co-evolution frameworks address this by expanding the training dataset with generated hard instances on which the current algorithm portfolio underperforms, thereby enhancing generalization. However, this paradigm faces two criti

저자 · Shaofeng Zhang, Shengcai Liu, Zhiyuan Wang, Ke Tang

arXiv에서 원문 보기