FunL2O Automates Feature Function Design for Learning-to-Optimize

FunL2O uses an LLM-driven program evolution loop to automate the design of feature functions for Learning-to-Optimize (L2O) pipelines. By evolving executable code and evaluating downstream performance on linear and quadratic programming tasks, it replaces manual feature engineering.