Advanced Genomes for Evolution of Artificial Neural Networks

Abstract

The thesis titled "Advanced Genomes for Evolution of Artificial Neural Networks" explores the integration of evolutionary algorithms with artificial neural networks (ANNs) to enhance their adaptability and efficiency. The research is motivated by the challenges associated with the scalability and complexity of evolving large neural networks, which traditional methods often struggle to address effectively. The study introduces a novel approach by incorporating hierarchical and modular genetic representations inspired by biological systems, particularly the Hox genes known for controlling the body plan of an organism during embryonic development. These genes provide a blueprint for modular and hierarchical organization, which can be analogous to structuring neural networks. By mimicking these biological patterns, the thesis aims to explore whether such a structured approach can lead to more efficient evolutionary outcomes compared to traditional methods. The primary objectives of this research are to investigate the effectiveness of hierarchical and modular genetic algorithms in evolving neural network architectures, compare the performance of neural networks evolved with Hox gene-inspired mutations against those evolved through traditional evolutionary methods, and analyze the impact of modular and hierarchical mutations on the adaptability and scalability of neural networks. This study is significant as it explores a novel approach to neural network evolution that could potentially reduce the computational overhead associated with traditional evolutionary algorithms. By integrating concepts from developmental biology into genetic algorithms, this research could pave the way for more sophisticated and efficient methods in neural network design and optimization. Furthermore, understanding the role of modularity and hierarchy in neural network evolution could provide deeper insights into both artificial and biological networks, enhancing our ability to design systems that are both robust and adaptable.