Optimizing neural network architectures for image recognition using genetic algorithms

This paper aims to present a method of implementing a better visual object recognition system with the inspiration gained from the processes of biological systems. Neural networks are closely related to biological systems in how they resemble the vertebra nervous system to perform classification tasks. However, in the success of neural networks, determining the configuration and the architecture of neural network plays a major role. Biological systems have evolved to their current state of cognition through natural evolution. Therefore, to attain an optimized neural network architecture for object recognition, the proposed system uses a genetic algorithm that simulates generations of neural network populations. A distributed parallel processing method is implemented on the system to undertake the enormous processing overhead required.