Pattern Recognition Capabilities of Neural Networks

Overview

Objectives/Goals The aim of the project was to experimentally determine the pattern recognition capabilities of neural networks by presenting them with mathematically generated input of varying complexity and consequently measuring their performance as the network complexity and learning parameters were modified. Methods/Materials A computer application developed by the author was used to perform the experiment. The application used the neural networks back-propagation learning algorithm to conduct testing. The exclusive or (XOR) operation was used in preliminary testing followed by multidimensional geometric figures, which were used as a source of patterns presented to the networks. Results Results of testing with XOR revealed a greater number of neurons leads to the need for a higher learning parameter in order to maintain a lower rate of error. The particular method used in the experiment was unable to successfully classify the multidimensional figures. The experiment revealed that error decreases proportionally to the learning rate and is unrelated to the number of neurons when the network is incapable of successfully handling a pattern. Conclusions/Discussion The overall information gleaned from the experiment suggests what forms of behavior researchers should expect when experimenting with neural networks (specifically the back-propagation algorithm).

Competition history

• CSEF 2003 Mathematics & Software · Entry S1227

Resources

• CSEF project summary ↗

Related projects

Closest projects by meaning, across every fair and year in the corpus.

Browse more like this

• All Computational Sciences projects
• CSEF Computational Sciences projects
• Every CSEF 2003 project