Skip to main navigation Skip to search Skip to main content

BOOST: Block Minifloat-Based On-Device CNN Training Accelerator with Transfer Learning

  • Chuliang Guo
  • , Binglei Lou
  • , Xueyuan Liu
  • , David Boland
  • , Philip H.W. Leong
  • , Cheng Zhuo*
  • *Corresponding author for this work
  • Zhejiang University
  • The University of Sydney
  • Key Lab of Cs&aus of Zhejiang Province

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

Abstract

Adapting CNNs to changing problems is challenging on resource-limited edge devices due to intensive computations, high precision requirements, large storage needs, and high bandwidth. This paper presents BOOST, a novel block minifloat (BM)-based parallel CNN training accelerator on memory- and computation-constrained FPGAs for transfer learning (TL). By updating a small number of layers online, BOOST enables adaptation to changing problems. Our approach utilizes a unified 8-bit BM datatype (bm(2,5) ), i.e., with a sign bit, 2 exponent bits, and 5 mantissa bits, and proposes unified Conv and dilated Conv blocks that support non-unit stride and enable task-level parallelism during back-propagation to minimize latency. For ResNet20 and VGG-like training on CIFAR-10 and SVHN datasets, BOOST achieves near 32-bit floating point accuracy, reducing latency by 21%-43% and BRAM usage by 63%-66% compared to back-propagation training without TL. Notably, BOOST outperforms the prior SOTA works to achieve perbatch throughput of 131 and 209 GOPs for ResNet20 and VGG-like respectively.

Original languageEnglish
Title of host publication2023 42nd IEEE/ACM International Conference on Computer-Aided Design, ICCAD 2023 - Proceedings
PublisherInstitute of Electrical and Electronics Engineers Inc.
ISBN (Electronic)9798350315592
DOIs
StatePublished - 2023
Externally publishedYes
Event42nd IEEE/ACM International Conference on Computer-Aided Design, ICCAD 2023 - San Francisco, United States
Duration: 28 Oct 20232 Nov 2023

Publication series

NameIEEE/ACM International Conference on Computer-Aided Design, Digest of Technical Papers, ICCAD
ISSN (Print)1092-3152

Conference

Conference42nd IEEE/ACM International Conference on Computer-Aided Design, ICCAD 2023
Country/TerritoryUnited States
CitySan Francisco
Period28/10/232/11/23

Fingerprint

Dive into the research topics of 'BOOST: Block Minifloat-Based On-Device CNN Training Accelerator with Transfer Learning'. Together they form a unique fingerprint.

Cite this