Mixed Precision DNNs: All you need is a good parametrization

TL;DR

This paper demonstrates that a well-chosen parametrization of the quantizer, specifically using step size and dynamic range, enables stable training and state-of-the-art mixed precision DNNs on CIFAR-10 and ImageNet.

Contribution

It introduces a novel parametrization of the quantizer for differentiable mixed precision training, improving stability and performance.

Findings

Parametrizing with step size and dynamic range is superior.

Achieved state-of-the-art results on CIFAR-10 and ImageNet.

Stable training of mixed precision networks with learned quantization.

Abstract

Efficient deep neural network (DNN) inference on mobile or embedded devices typically involves quantization of the network parameters and activations. In particular, mixed precision networks achieve better performance than networks with homogeneous bitwidth for the same size constraint. Since choosing the optimal bitwidths is not straight forward, training methods, which can learn them, are desirable. Differentiable quantization with straight-through gradients allows to learn the quantizer's parameters using gradient methods. We show that a suited parametrization of the quantizer is the key to achieve a stable training and a good final performance. Specifically, we propose to parametrize the quantizer with the step size and dynamic range. The bitwidth can then be inferred from them. Other parametrizations, which explicitly use the bitwidth, consistently perform worse. We confirm our findings with experiments on CIFAR-10 and ImageNet and we obtain mixed precision DNNs with learned quantization parameters, achieving state-of-the-art performance.