Tuning Pre-trained Model via Moment Probing

TL;DR

This paper introduces Moment Probing, a novel linear probing method that leverages feature distribution statistics for improved representation, and demonstrates its effectiveness across multiple benchmarks.

Contribution

The paper proposes Moment Probing, which exploits feature distribution moments for linear probing, and introduces MP+ with a partially shared module for enhanced fine-tuning.

Findings

MP outperforms traditional linear probing methods.

MP+ achieves state-of-the-art results on ten benchmarks.

Our approach requires less training cost than existing methods.

Abstract

Recently, efficient fine-tuning of large-scale pre-trained models has attracted increasing research interests, where linear probing (LP) as a fundamental module is involved in exploiting the final representations for task-dependent classification. However, most of the existing methods focus on how to effectively introduce a few of learnable parameters, and little work pays attention to the commonly used LP module. In this paper, we propose a novel Moment Probing (MP) method to further explore the potential of LP. Distinguished from LP which builds a linear classification head based on the mean of final features (e.g., word tokens for ViT) or classification tokens, our MP performs a linear classifier on feature distribution, which provides the stronger representation ability by exploiting richer statistical information inherent in features. Specifically, we represent feature distribution by its characteristic function, which is efficiently approximated by using first- and second-order moments of features. Furthermore, we propose a multi-head convolutional cross-covariance (MHC$^3$) to compute second-order moments in an efficient and effective manner. By considering that MP could affect feature learning, we introduce a partially shared module to learn two recalibrating parameters (PSRP) for backbones based on MP, namely MP$_{+}$. Extensive experiments on ten benchmarks using various models show that our MP significantly outperforms LP and is competitive with counterparts at less training cost, while our MP$_{+}$ achieves state-of-the-art performance.