Winograd Convolution: A Perspective from Fault Tolerance

TL;DR

This paper investigates the potential of Winograd convolution not only for computational efficiency but also for enhancing neural network fault tolerance, demonstrating significant reductions in design overhead and energy consumption.

Contribution

It is the first comprehensive evaluation of Winograd convolution's fault tolerance and explores its application for fault-tolerant and energy-efficient neural network processing.

Findings

Reduces fault-tolerant design overhead by 27.49%.

Lowers energy consumption by 7.19%.

Maintains accuracy without additional fault tolerance measures.

Abstract

Winograd convolution is originally proposed to reduce the computing overhead by converting multiplication in neural network (NN) with addition via linear transformation. Other than the computing efficiency, we observe its great potential in improving NN fault tolerance and evaluate its fault tolerance comprehensively for the first time. Then, we explore the use of fault tolerance of winograd convolution for either fault-tolerant or energy-efficient NN processing. According to our experiments, winograd convolution can be utilized to reduce fault-tolerant design overhead by 27.49\% or energy consumption by 7.19\% without any accuracy loss compared to that without being aware of the fault tolerance