Artificial DNA Framework for Autonomous Evolution of AI Models
DOI:
https://doi.org/10.54692/lgurjcsit.v10i1.199Keywords:
Artificial Intelligence, Graph Neural Networks, Machine Learning, Federated Learning, Artificial DNAAbstract
This research proposes an Artificial DNA (aDNA) framework to enable the autonomous evolution of AI models. Inspired by biological genetics, the framework encodes model architecture, hyperparameters, and learning behaviors into an artificial genome. A base model acts as a “parent,” whose aDNA is inherited, mutated, or recombined to generate “child” models. Each child undergoes performance evaluation on real-world tasks, and the fittest models are selected for the next generation. This continuous loop of inheritance, mutation, and selection fosters self-optimization and adaptability in evolving AI systems. The methodology ensures progressive model improvement without manual intervention. A case study on fraud detection demonstrates how the system adapts to new patterns by evolving its decision-making process. The proposed approach lays the groundwork for lifelong learning systems, enabling AI to evolve organically in complex environments.