Spatial Correlation and Value Prediction in Convolutional Neural Networks

TL;DR

This paper introduces a value prediction technique that leverages spatial correlation in CNNs to reduce multiply-accumulate operations by approximately 30%, with minimal accuracy loss, enhancing computational efficiency.

Contribution

The paper presents a novel value prediction method exploiting spatial correlation in CNN activations to significantly reduce computation without substantial accuracy loss.

Findings

Achieved 30.4% reduction in MAC operations on three CNNs for ImageNet.

Top-1 accuracy degraded by 1.7%, top-5 by 1.1%.

Method maintains high accuracy while improving efficiency.

Abstract

Convolutional neural networks (CNNs) are a widely used form of deep neural networks, introducing state-of-the-art results for different problems such as image classification, computer vision tasks, and speech recognition. However, CNNs are compute intensive, requiring billions of multiply-accumulate (MAC) operations per input. To reduce the number of MACs in CNNs, we propose a value prediction method that exploits the spatial correlation of zero-valued activations within the CNN output feature maps, thereby saving convolution operations. Our method reduces the number of MAC operations by 30.4%, averaged on three modern CNNs for ImageNet, with top-1 accuracy degradation of 1.7%, and top-5 accuracy degradation of 1.1%.