Evolutionary Artificial Neuroidal Network Using Serverless Architecture

Abstract

This thesis aims to address the limitations inherent in traditional Perceptron-based Artificial Neural Networks (ANNs), which are confined to linearly separable problems. An Evolutionary Artificial Neuroidal Network (EANN) is proposed to overcome these constraints, inspired by biological neural systems. Unlike the Perceptron, neuroids incorporate five adjustable parameters, offering greater flexibility and complexity in learning non-linear tasks. The research introduces a serverless cloud-based architecture to enable scalable and cost-efficient deployment of EANNs. Phase 1 implements an on-premise Model-View-Controller (MVC) three-tier architecture using Flask and Python for data processing and API management. Phase 2 transitions to the cloud, utilizing serverless AWS services, such as Lambda, Step Functions, and Batch to facilitate parallel processing and scalable model execution. A genetic algorithm is integrated to optimize iterative learning. Future work will further optimize cloud resource utilization, expand EANN’s applicability to complex AI domains, and advance real-time, data-driven health informatics applications.