Trade-offs in Neural Network Compression: Quantized and Binary Models for Keyword Spotting - LIRMM - Laboratoire d’Informatique, de Robotique et de Microélectronique de Montpellier
Conference Papers Year : 2024

Trade-offs in Neural Network Compression: Quantized and Binary Models for Keyword Spotting

Bruno Lovison-Franco
Jonathan Miquel
Aymen Romdhane
  • Function : Author
  • PersonId : 1421142
Lorena Anghel
David Novo
Pascal Benoit

Abstract

Enabling smart and independent IoT devices often requires to run complex Machine Learning (ML) workloads at the edge. Such systems usually operate with memories in the order of tens of kilobytes and low processing power. To fit within these constraints, model designers typically rely on low-precision integer representation of operations down to 1-bit, i.e., Binary Neural Networks (BNN). In this paper, we investigate the tradeoffs available to model designers between memory footprint and accuracy and the challenges to overcome for effective use of BNN. We show that designing BNN architectures is not a straightforward process. To overcome this, we propose a methodology based on design guidelines and Neural Architecture Search (NAS) to adapt traditional model architectures into BNN variants. As a case study, we apply this methodology to a ResNet-based model for a keyword spotting (KWS) application. Our results demonstrate that, contrary to 8-bit quantization, direct binarization significantly impacts accuracy. However, careful architecture redesign and hyperparameter tuning helps bringing BNNs performances on par with their quantized counterparts.
Fichier principal
Vignette du fichier
2024275754.pdf (312.67 Ko) Télécharger le fichier
Origin Files produced by the author(s)

Dates and versions

lirmm-04717703 , version 1 (02-10-2024)

Identifiers

  • HAL Id : lirmm-04717703 , version 1

Cite

Bruno Lovison-Franco, Jonathan Miquel, Aymen Romdhane, Guillaume Prenat, Lorena Anghel, et al.. Trade-offs in Neural Network Compression: Quantized and Binary Models for Keyword Spotting. ICECS 2024 - 31st IEEE International Conference on Electronics Circuits and Systems, Nov 2024, Nancy, France. In press. ⟨lirmm-04717703⟩
28 View
29 Download

Share

More