ARES-LSHADE: Autoresearch-Enhanced LSHADE with Memetic Polish for the GNBG Benchmark

In three linesARES-LSHADE, an LLM-designed differential-evolution variant, wins 510/744 function evaluations on the GNBG benchmark (GECCO 2026). Combines adaptive CMA-ES mutation and L-BFGS-B polish phase generated via 30 autoresearch experiments. Reaches machine precision on 18/24 functions; 6 remaining show characteristic plateaus. Code released.

Summary generated by Claude — human-verified