Domain-Aware Fine-Tuning: Enhancing Neural Network Adaptability

TL;DR

This paper introduces Domain-Aware Fine-Tuning (DAFT), a novel method that improves neural network adaptation by converting batch normalization layers and combining linear probing with fine-tuning, reducing feature distortion and enhancing performance.

Contribution

The paper proposes DAFT, a new fine-tuning approach that effectively mitigates feature distortion through batch normalization conversion and integrated linear probing, outperforming existing methods.

Findings

DAFT reduces feature distortion during fine-tuning.

DAFT improves performance on in-distribution and out-of-distribution datasets.

Extensive experiments show DAFT outperforms baseline methods.

Abstract

Fine-tuning pre-trained neural network models has become a widely adopted approach across various domains. However, it can lead to the distortion of pre-trained feature extractors that already possess strong generalization capabilities. Mitigating feature distortion during adaptation to new target domains is crucial. Recent studies have shown promising results in handling feature distortion by aligning the head layer on in-distribution datasets before performing fine-tuning. Nonetheless, a significant limitation arises from the treatment of batch normalization layers during fine-tuning, leading to suboptimal performance. In this paper, we propose Domain-Aware Fine-Tuning (DAFT), a novel approach that incorporates batch normalization conversion and the integration of linear probing and fine-tuning. Our batch normalization conversion method effectively mitigates feature distortion by reducing modifications to the neural network during fine-tuning. Additionally, we introduce the integration of linear probing and fine-tuning to optimize the head layer with gradual adaptation of the feature extractor. By leveraging batch normalization layers and integrating linear probing and fine-tuning, our DAFT significantly mitigates feature distortion and achieves improved model performance on both in-distribution and out-of-distribution datasets. Extensive experiments demonstrate that our method outperforms other baseline methods, demonstrating its effectiveness in not only improving performance but also mitigating feature distortion.