Evaluation of Early-exit Strategies in Low-cost FPGA-based Binarized Neural Networks

Minxuan Kong, Kris Nikov, Jose Luis Nunez-Yanez

Research output: Chapter in Book/Report/Conference proceedingConference Contribution (Conference Proceeding)

Abstract

In this paper, we investigate the application of early-exit strategies to quantized neural networks with binarized weights, mapped to low-cost FPGA SoC devices. The increasing complexity of network models means that hardware reuse and heterogeneous execution are needed and this opens the opportunity to evaluate the prediction confidence level early on. We apply the early-exit strategy to a network model suitable for ImageNet classification that combines weights with floating-point and binary arithmetic precision. The experiments show an improvement in inferred speed of around 20% using an early-exit network, compared with using a single primary neural network, with a negligible accuracy drop of 1.56%.

Original languageEnglish
Title of host publicationProceedings - 2022 25th Euromicro Conference on Digital System Design, DSD 2022
EditorsHimar Fabelo, Samuel Ortega, Amund Skavhaug
PublisherInstitute of Electrical and Electronics Engineers (IEEE)
Pages197-204
Number of pages8
ISBN (Electronic)9781665474047
DOIs
Publication statusPublished - 2022
Event25th Euromicro Conference on Digital System Design, DSD 2022 - Maspalomas, Spain
Duration: 31 Aug 20222 Sept 2022

Publication series

NameProceedings - 2022 25th Euromicro Conference on Digital System Design, DSD 2022

Conference

Conference25th Euromicro Conference on Digital System Design, DSD 2022
Country/TerritorySpain
CityMaspalomas
Period31/08/222/09/22

Bibliographical note

Funding Information:
ACKNOWLEDGMENTS This research was partially funded by the Royal Society Industry fellowship, INF\R2\192044 Machine Intelligence at the Network Edge (MINET), EPSRC HOPWARE EP\RV040863\1 and Leverhulme trust international fellowship High-performance video analytics with parallel heterogeneous neural networks IF-2021-003 .

Publisher Copyright:
© 2022 IEEE.

Keywords

  • Early-exit
  • FPGAs
  • Hardware Acceleration
  • Neural Network

Fingerprint

Dive into the research topics of 'Evaluation of Early-exit Strategies in Low-cost FPGA-based Binarized Neural Networks'. Together they form a unique fingerprint.

Cite this