Tag
This paper introduces DyCA, a framework for LLM-assisted evolutionary search that uses dynamic instance clustering to improve tail robustness under heterogeneous instance distributions, outperforming existing LES baselines on four algorithm design tasks.