Surgical Fine-Tuning Improves Adaptation to Distribution Shifts

TL;DR

This paper introduces surgical fine-tuning, a selective approach to adapt pre-trained models to distribution shifts, demonstrating its effectiveness across various tasks and providing theoretical insights into its advantages.

Contribution

It proposes surgical fine-tuning, showing that selectively tuning layers can outperform full fine-tuning, with theoretical backing for certain neural network settings.

Findings

Selective fine-tuning matches or outperforms standard methods.

Effectiveness depends on the type of distribution shift.

Theoretical proof for first-layer tuning superiority in idealized models.

Abstract

A common approach to transfer learning under distribution shift is to fine-tune the last few layers of a pre-trained model, preserving learned features while also adapting to the new task. This paper shows that in such settings, selectively fine-tuning a subset of layers (which we term surgical fine-tuning) matches or outperforms commonly used fine-tuning approaches. Moreover, the type of distribution shift influences which subset is more effective to tune: for example, for image corruptions, fine-tuning only the first few layers works best. We validate our findings systematically across seven real-world data tasks spanning three types of distribution shifts. Theoretically, we prove that for two-layer neural networks in an idealized setting, first-layer tuning can outperform fine-tuning all layers. Intuitively, fine-tuning more parameters on a small target dataset can cause information learned during pre-training to be forgotten, and the relevant information depends on the type of shift.