Skip to Main content Skip to Navigation
Book sections

Cross-layer CNN Approximations for Hardware Implementation

Abstract : 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.
Document type :
Book sections
Complete list of metadata

https://hal-uphf.archives-ouvertes.fr/hal-03677575
Contributor : Mylène Delrue Connect in order to contact the contributor
Submitted on : Tuesday, May 24, 2022 - 5:08:42 PM
Last modification on : Wednesday, May 25, 2022 - 3:42:25 AM

Identifiers

Citation

Ali Karimi, Ihsen Alouani, Abdessamad Ait El Cadi, Hamza Ouarnoughi, Smail Niar. Cross-layer CNN Approximations for Hardware Implementation. Applied Reconfigurable Computing. Architectures, Tools, and Applications, 12083, Springer International Publishing, pp.151-165, 2020, Lecture Notes in Computer Science, 978-3-030-44533-1. ⟨10.1007/978-3-030-44534-8_12⟩. ⟨hal-03677575⟩

Share

Metrics

Record views

2