Deep Learning with Limited Numerical Precision

TL;DR

This paper demonstrates that deep neural networks can be effectively trained using 16-bit fixed-point numbers with stochastic rounding, enabling energy-efficient hardware implementations without significant accuracy loss.

Contribution

It introduces the use of stochastic rounding in low-precision fixed-point training and presents a hardware accelerator for energy-efficient neural network training.

Findings

Deep networks trained with 16-bit fixed-point and stochastic rounding maintain accuracy.

Stochastic rounding significantly improves training stability in low-precision arithmetic.

Hardware accelerator demonstrates energy-efficient fixed-point neural network training.

Abstract

Training of large-scale deep neural networks is often constrained by the available computational resources. We study the effect of limited precision data representation and computation on neural network training. Within the context of low-precision fixed-point computations, we observe the rounding scheme to play a crucial role in determining the network's behavior during training. Our results show that deep networks can be trained using only 16-bit wide fixed-point number representation when using stochastic rounding, and incur little to no degradation in the classification accuracy. We also demonstrate an energy-efficient hardware accelerator that implements low-precision fixed-point arithmetic with stochastic rounding.