Efficient Low-rank Backpropagation for Vision Transformer Adaptation

TL;DR

This paper introduces LBP-WHT, a low-rank backpropagation method using Walsh-Hadamard transformation, significantly reducing computation in ViT fine-tuning while improving accuracy on datasets like CIFAR100.

Contribution

We propose the first low-rank backpropagation technique for accelerating vision transformer adaptation using Walsh-Hadamard transformation.

Findings

Achieves 10.4% higher accuracy on CIFAR100 with less computation.

Reduces backpropagation computation by 9 MFLOPs.

Effective across different ViT and hybrid models.

Abstract

The increasing scale of vision transformers (ViT) has made the efficient fine-tuning of these large models for specific needs a significant challenge in various applications. This issue originates from the computationally demanding matrix multiplications required during the backpropagation process through linear layers in ViT. In this paper, we tackle this problem by proposing a new Low-rank BackPropagation via Walsh-Hadamard Transformation (LBP-WHT) method. Intuitively, LBP-WHT projects the gradient into a low-rank space and carries out backpropagation. This approach substantially reduces the computation needed for adapting ViT, as matrix multiplication in the low-rank space is far less resource-intensive. We conduct extensive experiments with different models (ViT, hybrid convolution-ViT model) on multiple datasets to demonstrate the effectiveness of our method. For instance, when adapting an EfficientFormer-L1 model on CIFAR100, our LBP-WHT achieves 10.4% higher accuracy than the state-of-the-art baseline, while requiring 9 MFLOPs less computation. As the first work to accelerate ViT adaptation with low-rank backpropagation, our LBP-WHT method is complementary to many prior efforts and can be combined with them for better performance.