Robust fine-tuning of zero-shot models

TL;DR

The paper introduces WiSE-FT, a simple ensembling method that enhances the robustness of fine-tuned zero-shot models against distribution shifts without sacrificing target accuracy.

Contribution

It proposes WiSE-FT, an ensembling technique that improves robustness during fine-tuning of large models like CLIP, maintaining high accuracy across diverse data distributions.

Findings

WiSE-FT improves accuracy under distribution shift by 4-6 percentage points.

WiSE-FT increases ImageNet accuracy by 1.6 percentage points.

WiSE-FT achieves large robustness gains across multiple datasets.

Abstract

Large pre-trained models such as CLIP or ALIGN offer consistent accuracy across a range of data distributions when performing zero-shot inference (i.e., without fine-tuning on a specific dataset). Although existing fine-tuning methods substantially improve accuracy on a given target distribution, they often reduce robustness to distribution shifts. We address this tension by introducing a simple and effective method for improving robustness while fine-tuning: ensembling the weights of the zero-shot and fine-tuned models (WiSE-FT). Compared to standard fine-tuning, WiSE-FT provides large accuracy improvements under distribution shift, while preserving high accuracy on the target distribution. On ImageNet and five derived distribution shifts, WiSE-FT improves accuracy under distribution shift by 4 to 6 percentage points (pp) over prior work while increasing ImageNet accuracy by 1.6 pp. WiSE-FT achieves similarly large robustness gains (2 to 23 pp) on a diverse set of six further distribution shifts, and accuracy gains of 0.8 to 3.3 pp compared to standard fine-tuning on seven commonly used transfer learning datasets. These improvements come at no additional computational cost during fine-tuning or inference.