Training Artificial Immune Networks as Standalone Generative Models for Realistic Data Synthesis

This chapter delves into the innovative application of Artificial Immune Networks (AINs) as standalone generative models for realistic data synthesis. It begins by introducing Generative Artificial Intelligence (GAI) and its subfield, Generative Modelling, which aims to model data distributions to generate human-like content. The chapter then explores various techniques for Generative Modelling, including Maximum Likelihood Estimation, Boltzmann Machines, and Variational Autoencoders. The focus shifts to Generative Adversarial Networks (GANs), highlighting their success and the challenges they face, such as vanishing gradients and mode collapse. The core of the chapter introduces the concept of Artificial Immune Networks and their application in generative models. It presents the GAAINet prototype, which trains an AIN to learn a dataset's distribution and generate synthetic samples. The chapter provides a detailed overview of the GAAINet framework, including the generator and discriminator agents, their training processes, and the experimental results obtained. It compares the performance of GAAINet to first-generation GANs, showcasing the potential of AINs in generating human-readable synthetic samples. The chapter concludes by discussing future research directions and the promise of AIN-based generative models.