2025 ICML ICML 2025

Adapting to Linear Separable Subsets with Large-Margin in Differentially Private Learning