Is the input layer of an artificial neural network fully-convolving the input data with the weights matrix? An accurate architectural model of a neural network as a correlator
-
1
University of Sussex, Dept. of Engineering and Design, United Kingdom
Previous research work in literature has focused on understanding and modeling the internal functions of an artificial neural network for engineering applications. Several models before have described the architecture of a feedforward backpropagation neural network (see Fig. 1) by a correlators-type network. In our study presented here, we prove that such a model is not, in general, an accurate representation of a neural network. To help us in our proof, we investigate two different approaches of inputting the data to the neural network for image processing engineering problems; namely by lexicographical scanning of the data matrix or by inputting it as a two-dimensional matrix similar to an optical correlator. Then, we further study in detail the first approach of the lexicographical scanning by mathematically analyzing the two possible ways of row-by-row lexicographical scanning and raster lexicographical scanning. We employ the convolution theorem and compare both approaches of the lexicographical scanning and of the two-dimensional matrix input to the neural network of the same image data. From the conclusions extracted from our study, we propose a new architectural model (see Fig. 2) for now accurately describing a neural network as a correlator.
Conference:
Neuroinformatics 2009, Pilsen, Czechia, 6 Sep - 8 Sep, 2009.
Presentation Type:
Poster Presentation
Topic:
General neuroinformatics
Citation:
Kypraios
I
(2019). Is the input layer of an artificial neural network fully-convolving the input data with the weights matrix? An accurate architectural model of a neural network as a correlator.
Front. Neuroinform.
Conference Abstract:
Neuroinformatics 2009.
doi: 10.3389/conf.neuro.11.2009.08.085
Copyright:
The abstracts in this collection have not been subject to any Frontiers peer review or checks, and are not endorsed by Frontiers.
They are made available through the Frontiers publishing platform as a service to conference organizers and presenters.
The copyright in the individual abstracts is owned by the author of each abstract or his/her employer unless otherwise stated.
Each abstract, as well as the collection of abstracts, are published under a Creative Commons CC-BY 4.0 (attribution) licence (https://creativecommons.org/licenses/by/4.0/) and may thus be reproduced, translated, adapted and be the subject of derivative works provided the authors and Frontiers are attributed.
For Frontiers’ terms and conditions please see https://www.frontiersin.org/legal/terms-and-conditions.
Received:
22 May 2009;
Published Online:
09 May 2019.
*
Correspondence:
Ioannis Kypraios, University of Sussex, Dept. of Engineering and Design, Sussex, United Kingdom, i.kypraios@sussex.ac.uk