ABSTRACT
This paper introduces an AI-Based Distributed Model Building System that uses Large Language Models (LLMs) and distributed AI agents to autonomously design, evaluate, and optimize Artificial Neural Networks (ANNs). The system iterates over 10+ cycles, exploring and ranking architectures based on performance, refining them with a 2:8 good-to-bad ratio. A dynamic rule base evolves with decay mechanisms for efficient exploration, while techniques like Importance Matrix Analysis and Modal Value Aggregation identify key layers and robust hyperparameters. The final ANN achieves 95% accuracy in binary classification tasks. Distributed processing enhances scalability and efficiency, making this system suitable for model building.
