Trained Quantization Thresholds for Accurate and Efficient Fixed-Point Inference of Deep Neural Networks

TL;DR

This paper introduces a training method for quantization thresholds in neural networks that improves accuracy and efficiency, enabling near-floating-point performance with minimal retraining, and provides a framework for automatic quantization.

Contribution

The paper presents a novel training approach for quantization thresholds using backpropagation, with analytical insights and hardware-friendly constraints, achieving high accuracy with minimal retraining.

Findings

Achieves near-floating-point accuracy on MobileNets with less than 5 epochs of retraining.

Uses power-of-2 scale-factors and per-tensor scaling for hardware efficiency.

Provides a framework for automatic TensorFlow graph quantization.

Abstract

We propose a method of training quantization thresholds (TQT) for uniform symmetric quantizers using standard backpropagation and gradient descent. Contrary to prior work, we show that a careful analysis of the straight-through estimator for threshold gradients allows for a natural range-precision trade-off leading to better optima. Our quantizers are constrained to use power-of-2 scale-factors and per-tensor scaling of weights and activations to make it amenable for hardware implementations. We present analytical support for the general robustness of our methods and empirically validate them on various CNNs for ImageNet classification. We are able to achieve near-floating-point accuracy on traditionally difficult networks such as MobileNets with less than 5 epochs of quantized (8-bit) retraining. Finally, we present Graffitist, a framework that enables automatic quantization of TensorFlow graphs for TQT (available at https://github.com/Xilinx/graffitist ).