DL-CNN: Double Layered Convolutional Neural Networks

Lixin Fu, Rohith Rangineni

2022

Abstract

We studied the traditional convolutional neural networks and developed a new model that used double layers instead of only one. In our example of this model, we used five convolutional layers and four fully connected layers. The dataset has four thousand human face images of two classes, one of them being open eyes and the other closed eyes. In this project, we dissected the original source code of the standard package into several components and changed some of the core parts to improve accuracy. In addition to using both the current layer and the prior layer to compute the next layer, we also explored whether to skip the current layer. We changed the original convolution window formula. A multiplication bias instead of originally adding bias to the linear combination was also proposed. Though it is hard to explain the rationale, the results of multiplication bias are better in our example. For our new double layer model, our simulation results showed that the accuracy was increased from 60% to 95%.

Paper Citation

in Harvard Style

Fu L. and Rangineni R. (2022). DL-CNN: Double Layered Convolutional Neural Networks. In Proceedings of the 24th International Conference on Enterprise Information Systems - Volume 2: ICEIS, ISBN 978-989-758-569-2, pages 281-286. DOI: 10.5220/0011117000003179

in Bibtex Style

@conference{iceis22,
author={Lixin Fu and Rohith Rangineni},
title={DL-CNN: Double Layered Convolutional Neural Networks},
booktitle={Proceedings of the 24th International Conference on Enterprise Information Systems - Volume 2: ICEIS,},
year={2022},
pages={281-286},
publisher={SciTePress},
organization={INSTICC},
doi={10.5220/0011117000003179},
isbn={978-989-758-569-2},
}

in EndNote Style

TY - CONF

JO - Proceedings of the 24th International Conference on Enterprise Information Systems - Volume 2: ICEIS,
TI - DL-CNN: Double Layered Convolutional Neural Networks
SN - 978-989-758-569-2
AU - Fu L.
AU - Rangineni R.
PY - 2022
SP - 281
EP - 286
DO - 10.5220/0011117000003179