Cross-layer CNN Approximations for Hardware ...
Type de document :
Partie d'ouvrage
Titre :
Cross-layer CNN Approximations for Hardware Implementation
Auteur(s) :
Karimi, Ali [Auteur]
Laboratoire d'Automatique, de Mécanique et d'Informatique industrielles et Humaines - UMR 8201 [LAMIH]
Alouani, Lihsen [Auteur]
Institut d’Électronique, de Microélectronique et de Nanotechnologie - UMR 8520 [IEMN]
COMmunications NUMériques - IEMN [COMNUM - IEMN]
INSA Institut National des Sciences Appliquées Hauts-de-France [INSA Hauts-De-France]
Ait El Cadi, Abdessamad [Auteur]
Laboratoire d'Automatique, de Mécanique et d'Informatique industrielles et Humaines - UMR 8201 [LAMIH]
Ouarnoughi, Hamza [Auteur]
Laboratoire d'Automatique, de Mécanique et d'Informatique industrielles et Humaines - UMR 8201 [LAMIH]
Niar, Smail [Auteur]
Laboratoire d'Automatique, de Mécanique et d'Informatique industrielles et Humaines - UMR 8201 [LAMIH]
Laboratoire d'Automatique, de Mécanique et d'Informatique industrielles et Humaines - UMR 8201 [LAMIH]
Alouani, Lihsen [Auteur]
Institut d’Électronique, de Microélectronique et de Nanotechnologie - UMR 8520 [IEMN]
COMmunications NUMériques - IEMN [COMNUM - IEMN]
INSA Institut National des Sciences Appliquées Hauts-de-France [INSA Hauts-De-France]
Ait El Cadi, Abdessamad [Auteur]
Laboratoire d'Automatique, de Mécanique et d'Informatique industrielles et Humaines - UMR 8201 [LAMIH]
Ouarnoughi, Hamza [Auteur]
Laboratoire d'Automatique, de Mécanique et d'Informatique industrielles et Humaines - UMR 8201 [LAMIH]
Niar, Smail [Auteur]
Laboratoire d'Automatique, de Mécanique et d'Informatique industrielles et Humaines - UMR 8201 [LAMIH]
Titre de l’ouvrage :
Applied Reconfigurable Computing. Architectures, Tools, and Applications
Éditeur :
Springer International Publishing
Lieu de publication :
Cham
Date de publication :
2020-03-25
ISBN :
978-3-030-44533-1
Mot(s)-clé(s) en anglais :
CNNs
FPGA
Approximate computing
FPGA
Approximate computing
Discipline(s) HAL :
Informatique [cs]
Résumé en anglais : [en]
Convolution Neural Networks (CNNs) are widely used for image classification and object detection applications. The deployment of these architectures in embedded applications is a great challenge. This challenge arises from ...
Lire la suite >Convolution Neural Networks (CNNs) are widely used for image classification and object detection applications. The deployment of these architectures in embedded applications is a great challenge. This challenge arises from CNNs’ high computation complexity that is required to be implemented on platforms with limited hardware resources like FPGA. Since these applications are inherently error-resilient, approximate computing (AC) offers an interesting trade-off between resource utilization and accuracy. In this paper, we study the impact on CNN performances when several approximation techniques are applied simultaneously. We focus on two of the widely used approximation techniques, namely quantization and pruning. Our experimental results showed that for CNN networks of different parameter sizes and 3% loss in accuracy, we can obtain up to 27.9%–47.2% reduction in computation complexity in terms of FLOPs for CIFAR-10 and MNIST datasets.Lire moins >
Lire la suite >Convolution Neural Networks (CNNs) are widely used for image classification and object detection applications. The deployment of these architectures in embedded applications is a great challenge. This challenge arises from CNNs’ high computation complexity that is required to be implemented on platforms with limited hardware resources like FPGA. Since these applications are inherently error-resilient, approximate computing (AC) offers an interesting trade-off between resource utilization and accuracy. In this paper, we study the impact on CNN performances when several approximation techniques are applied simultaneously. We focus on two of the widely used approximation techniques, namely quantization and pruning. Our experimental results showed that for CNN networks of different parameter sizes and 3% loss in accuracy, we can obtain up to 27.9%–47.2% reduction in computation complexity in terms of FLOPs for CIFAR-10 and MNIST datasets.Lire moins >
Langue :
Anglais
Audience :
Internationale
Vulgarisation :
Non
Source :
Fichiers
- ARC_2020.pdf
- Accès libre
- Accéder au document