Bridge deep learning to the physical world: An efficient method to quantize network

Pei Hen Hung, Chia-Han Lee, Shao Wen Yang, V. Srinivasa Somayazulu, Yen Kuang Chen, Shao Yi Chien

Research output: Chapter in Book/Report/Conference proceedingConference contributionpeer-review

8 Scopus citations

Abstract

As better performance is achieved by deep convolutional network with more and more layers, the increasing number of weighting and bias parameters makes it only possible to be implemented on servers in cyber space but infeasible to be deployed in physical-world embedded systems because of huge storage and memory bandwidth requirements. In this paper, we proposed an efficient method to quantize the model parameters. Instead of taking the quantization process as a negative effect on precision, we regarded it as a regularize problem to prevent overfitting, and a two-stage quantization technique including soft- and hard-quantization is developed. With the help of our quantization method, not only 93.75% of the parameter memory size can be reduced by replacing the word length from 32-bit to 2-bit, but the testing accuracy after quantization is also better than previous approaches in some dataset, and the additional training overhead is only 3% of the ordinary one.

Original languageEnglish
Title of host publicationElectronic Proceedings of the 2015 IEEE International Workshop on Signal Processing Systems, SiPS 2015
PublisherInstitute of Electrical and Electronics Engineers Inc.
ISBN (Electronic)9781467396042
DOIs
StatePublished - 2 Dec 2015
EventIEEE International Workshop on Signal Processing Systems, SiPS 2015 - Hangzhou, China
Duration: 14 Oct 201516 Oct 2015

Publication series

NameIEEE Workshop on Signal Processing Systems, SiPS: Design and Implementation
Volume2015-December
ISSN (Print)1520-6130

Conference

ConferenceIEEE International Workshop on Signal Processing Systems, SiPS 2015
CountryChina
CityHangzhou
Period14/10/1516/10/15

Keywords

  • Convolutional Network
  • Deep Learning
  • Deep Neural Network
  • Quantize
  • Regularize

Fingerprint Dive into the research topics of 'Bridge deep learning to the physical world: An efficient method to quantize network'. Together they form a unique fingerprint.

Cite this