Data-Free Quantization Through Weight Equalization and Bias Correction (2019-06-11T00:00:00.000000Z)

TL;DR

This work introduces a data-free quantization method for deep neural networks that does not require fine-tuning or hyperparameter selection, and achieves near-original model performance on common computer vision architectures and tasks.

Abstract

We introduce a data-free quantization method for deep neural networks that does not require fine-tuning or hyperparameter selection. It achieves near-original model performance on common computer vision architectures and tasks. 8-bit fixed-point quantization is essential for efficient inference on modern deep learning hardware. However, quantizing models to run in 8-bit is a non-trivial task, frequently leading to either significant performance reduction or engineering time spent on training a network to be amenable to quantization. Our approach relies on equalizing the weight ranges in the network by making use of a scale-equivariance property of activation functions. In addition the method corrects biases in the error that are introduced during quantization. This improves quantization accuracy performance, and can be applied to many common computer vision architectures with a straight forward API call. For common architectures, such as the MobileNet family, we achieve state-of-the-art quantized model performance. We further show that the method also extends to other computer vision architectures and tasks such as semantic segmentation and object detection.

Authors

M. Welling

17 papers

Markus Nagel

1 papers

M. V. Baalen

1 papers

TL;DR

Abstract

Authors

References39 items

Fighting Quantization Bias With Bias

Equi-normalization of Neural Networks

Same, Same But Different - Recovering Neural Network Quantization Error Through Weight Factorization

Probabilistic Binary Neural Networks

Quantizing deep convolutional networks for efficient inference: A whitepaper

SYQ: Learning Symmetric Quantization for Efficient Deep Neural Networks

Ristretto: A Framework for Empirical Study of Resource-Efficient Inference in Convolutional Neural Networks

A Quantization-Friendly Separable Convolution for MobileNets

PACT: Parameterized Clipping Activation for Quantized Neural Networks

Variational Network Quantization

Encoder-Decoder with Atrous Separable Convolution for Semantic Image Segmentation

MobileNetV2: Inverted Residuals and Linear Bottlenecks

Quantization and Training of Neural Networks for Efficient Integer-Arithmetic-Only Inference

Flexpoint: An Adaptive Numerical Format for Efficient Training of Deep Neural Networks

Automatic differentiation in PyTorch

Channel Pruning for Accelerating Very Deep Neural Networks

ShiftCNN: Generalized Low-Precision Architecture for Inference of Convolutional Neural Networks

MobileNets: Efficient Convolutional Neural Networks for Mobile Vision Applications

WRPN: Training and Inference using Wide Reduced-Precision Networks

Soft Weight-Sharing for Neural Network Compression

Incremental Network Quantization: Towards Lossless CNNs with Low-Precision Weights

Quantized Neural Networks: Training Neural Networks with Low Precision Weights and Activations

On the Efficient Representation and Execution of Deep Acoustic Models

DoReFa-Net: Training Low Bitwidth Convolutional Neural Networks with Low Bitwidth Gradients

XNOR-Net: ImageNet Classification Using Binary Convolutional Neural Networks

Convolutional Neural Networks using Logarithmic Data Representation

Deep Residual Learning for Image Recognition

SSD: Single Shot MultiBox Detector

BinaryConnect: Training Deep Neural Networks with binary weights during propagations

Accelerating Very Deep Convolutional Networks for Classification and Detection

Deep Learning with Limited Numerical Precision

Delving Deep into Rectifiers: Surpassing Human-Level Performance on ImageNet Classification

ImageNet Large Scale Visual Recognition Challenge

The Pascal Visual Object Classes Challenge: A Retrospective

Rectified Linear Units Improve Restricted Boltzmann Machines

27th International Conference on Machine Learning (ICML-10)

UvA-DARE (Digital Academic Repository) Relaxed Quantization for Discretized Neural Networks R ELAXED Q UANTIZATION FOR D ISCRETIZED N EURAL N ETWORKS

Level 2 Requires data but no backpropagation. Works for any model. The data

Level 1 No data and no backpropagation required. Method works for any model. As simple as an API call that only looks at the model deﬁnition and weights

Field of Study

Journal Information

Name

Page

Venue Information

Name

Type

URL

Alternate Names