Training Acceleration of Low-Rank Decomposed Networks using Sequential Freezing and Rank Quantization

TL;DR

This paper introduces techniques to accelerate low-rank decomposed neural networks by optimizing ranks and freezing layers sequentially, significantly improving training and inference speed while maintaining accuracy.

Contribution

The paper proposes rank optimization and sequential freezing methods to enhance training acceleration of low-rank decomposed models without reducing decomposition ranks.

Findings

Up to 60% training throughput improvement

Up to 37% inference speedup

Maintains accuracy close to original models

Abstract

Low Rank Decomposition (LRD) is a model compression technique applied to the weight tensors of deep learning models in order to reduce the number of trainable parameters and computational complexity. However, due to high number of new layers added to the architecture after applying LRD, it may not lead to a high training/inference acceleration if the decomposition ranks are not small enough. The issue is that using small ranks increases the risk of significant accuracy drop after decomposition. In this paper, we propose two techniques for accelerating low rank decomposed models without requiring to use small ranks for decomposition. These methods include rank optimization and sequential freezing of decomposed layers. We perform experiments on both convolutional and transformer-based models. Experiments show that these techniques can improve the model throughput up to 60% during training and 37% during inference when combined together while preserving the accuracy close to that of the original models