The Effects of Fully Connected Layers Adjustment for Lightweight Convolutional Neural Networks

Reducing the size and the computational need of a Convolutional Neural Network (CNN) is a crucial task that can be extremely important for the deployment phase of such models. In this work, we are quoting the most efficient techniques in order to reduce the size of CNNs while maintaining prediction...

Full description

Saved in:
Bibliographic Details
Published inInternational Conference on Innovation and Intelligence for Informatics, Computing, and Technologies (Online) pp. 50 - 57
Main Authors Nerantzis, Evangelos, Kazakis, Apostolos, Symeonidis, Georgios, Papakostas, George A.
Format Conference Proceeding
LanguageEnglish
Published IEEE 20.11.2022
Subjects
CNN
compression techniques
computer vision
deep learning
IoT
Lightweight
Online AccessGet full text
ISSN2770-7466
DOI10.1109/3ICT56508.2022.9990841

Cover

More Information
Summary:Reducing the size and the computational need of a Convolutional Neural Network (CNN) is a crucial task that can be extremely important for the deployment phase of such models. In this work, we are quoting the most efficient techniques in order to reduce the size of CNNs while maintaining prediction accuracy. Through a case study, we investigate the effects, especially of the dense layers and how the number of nodes in each layer can change the weight, performance and training process of a CNN. Although there are lots of works around this subject, in our paper we are proving how and why the number of nodes in fully connected (dense) layers is the most important factor for reducing the size of a CNN.
ISSN:2770-7466
DOI:10.1109/3ICT56508.2022.9990841