MEC: Memory-efficient Convolution for Deep Neural Network

TL;DR

This paper introduces MEC, a memory-efficient convolution method that reduces memory overhead and accelerates deep neural network convolution operations by using compact lowering and parallel small matrix multiplications.

Contribution

MEC is a novel convolution algorithm that significantly reduces memory usage and improves speed compared to existing indirect methods.

Findings

MEC reduces memory consumption substantially.

MEC achieves good speedup on mobile and server platforms.

MEC outperforms other indirect convolution algorithms in efficiency.

Abstract

Convolution is a critical component in modern deep neural networks, thus several algorithms for convolution have been developed. Direct convolution is simple but suffers from poor performance. As an alternative, multiple indirect methods have been proposed including im2col-based convolution, FFT-based convolution, or Winograd-based algorithm. However, all these indirect methods have high memory-overhead, which creates performance degradation and offers a poor trade-off between performance and memory consumption. In this work, we propose a memory-efficient convolution or MEC with compact lowering, which reduces memory-overhead substantially and accelerates convolution process. MEC lowers the input matrix in a simple yet efficient/compact way (i.e., much less memory-overhead), and then executes multiple small matrix multiplications in parallel to get convolution completed. Additionally, the reduced memory footprint improves memory sub-system efficiency, improving performance. Our experimental results show that MEC reduces memory consumption significantly with good speedup on both mobile and server platforms, compared with other indirect convolution algorithms.