Dual Risk Minimization: Towards Next-Level Robustness in Fine-tuning Zero-Shot Models

TL;DR

This paper introduces dual risk minimization (DRM), a novel fine-tuning method that balances expected and worst-case risks to enhance robustness of foundation models against distribution shifts, achieving state-of-the-art results.

Contribution

The paper proposes DRM, a new fine-tuning approach that leverages core features and worst-case risk estimation to improve model robustness beyond existing methods.

Findings

DRM improves out-of-distribution accuracy on multiple benchmarks.

DRM achieves state-of-the-art performance in robustness tasks.

Utilizes core-feature descriptions from LLMs for risk estimation.

Abstract

Fine-tuning foundation models often compromises their robustness to distribution shifts. To remedy this, most robust fine-tuning methods aim to preserve the pre-trained features. However, not all pre-trained features are robust and those methods are largely indifferent to which ones to preserve. We propose dual risk minimization (DRM), which combines empirical risk minimization with worst-case risk minimization, to better preserve the core features of downstream tasks. In particular, we utilize core-feature descriptions generated by LLMs to induce core-based zero-shot predictions which then serve as proxies to estimate the worst-case risk. DRM balances two crucial aspects of model robustness: expected performance and worst-case performance, establishing a new state of the art on various real-world benchmarks. DRM significantly improves the out-of-distribution performance of CLIP ViT-L/14@336 on ImageNet (75.9 to 77.1), WILDS-iWildCam (47.1 to 51.8), and WILDS-FMoW (50.7 to 53.1); opening up new avenues for robust fine-tuning. Our code is available at https://github.com/vaynexie/DRM .