Transform-Based Feature Map Compression for CNN Inference

TL;DR

This paper introduces DCT-CM, a transform-based method that exploits layer-specific sparsity to significantly improve feature map compression in CNNs, reducing memory bandwidth and power consumption during inference.

Contribution

It presents a novel hardware-friendly transform method that leverages layer sparsity distinctions, achieving higher compression ratios than existing approaches.

Findings

Achieves 2.9x average compression ratio on ResNet-50

53% higher compression ratio than previous transform-based methods

Effective during inference with 8-bit quantization

Abstract

To achieve higher accuracy in machine learning tasks, very deep convolutional neural networks (CNNs) are designed recently. However, the large memory access of deep CNNs will lead to high power consumption. A variety of hardware-friendly compression methods have been proposed to reduce the data transfer bandwidth by exploiting the sparsity of feature maps. Most of them focus on designing a specialized encoding format to increase the compression ratio. Differently, we observe and exploit the sparsity distinction between activations in earlier and later layers to improve the compression ratio. We propose a novel hardware-friendly transform-based method named 1D-Discrete Cosine Transform on Channel dimension with Masks (DCT-CM), which intelligently combines DCT, masks, and a coding format to compress activations. The proposed algorithm achieves an average compression ratio of 2.9x (53% higher than the state-of-the-art transform-based feature map compression works) during inference on ResNet-50 with an 8-bit quantization scheme.