Optimized Conformal Selection: Powerful Selective Inference After Data-Dependent Score Construction
Optimized Conformal Selection (OptCS) is a framework that allows for valid statistical model selection after flexible, data-driven optimization. It introduces three procedures—OptCS-MSel, OptCS-Full, and OptCS-Full-MSel—to maintain finite-sample False Discovery Rate control while maximizing predictive power by intelligently selecting models or using all available labeled data.