Home // SIGNAL 2025, The Tenth International Conference on Advances in Signal, Image and Video Processing // View article

Efficient Implementation of CNN in Deep Learning by Using Multirate Algorithms

Authors:
Guowei Xiao
Yingshuai Wang
Ping Wang

Keywords: CNN, ML, DL, AI, IC, Multirate, 2-D, Signal Processing, DSP, AISC, Filter

Abstract:
This paper proposes the multirate Convolutional Neural Networks (CNN) algorithms for an efficient implementation of the 2-Dimensional (2-D) CNN circuits implementation. During the rapid growth in computation power, Deep Learning (DL) using CNN has widened the areas of the Artificial Intelligent (AI) applications. For the layers of the convolution with pooling operation in CNN some researchers work has initially applied the multirate algorithms to the traditional (non-multirate) convolutional kernel operation of using polyphase architectures resulting in the more efficient implementation of the multirate filtering. In this work we extend it into 2-D CNN by using time-varying coefficient to achieve an efficient implementation with reduced memory(i.e. the line-buffer) size by M-fold(the pooling factor) and the MACs at 1/M of clock running rate. A design example of the first stage of CNN system will be provided. Its results are verified with the Matlab CNN-based digit recognition tool.

Pages: 28 to 32

Copyright: Copyright (c) IARIA, 2025

Publication date: March 9, 2025

Published in: conference

ISSN: 2519-8432

ISBN: 978-1-68558-245-6

Location: Lisbon, Portugal

Dates: from March 9, 2025 to March 13, 2025