Learning the Robustness Mechanism with Bilevel Optimization

The paper introduces a distributionally robust learning framework that learns the robustness mechanism’s parameters from held‑out data via bilevel optimization, avoiding extensive manual tuning. Two variants of the framework are presented: one for settings with group labels and one for settings without them. The authors provide theoretical sample‑complexity guarantees comparable to exhaustive grid search and demonstrate empirically that the method scales and performs well when both intra‑group and inter‑group distribution shifts occur simultaneously.

Machine-generated by The Flow from the publisher's headline and feed description — not written or checked by a human. The full article lives at Hugging Face Trending Papers.