Generic distributed polymorphic learning model for a community of heterogeneous cyber physical social robots in MAS Environment and GPU Architecture

Abstract

This paper presents a new distributed polymorphic learning model for a community of heterogeneous cyber physical robots operating in a multi agent environment. This model allows a community of intelligent physical agents to exchange their minds represented by configured and trained neural net-works. The training operation of the neural networks is performed, using machines and deep learning techniques, in a distributed way based on special agents deployed in machines having high-performance computing resources based on GPUs. Each mind, specialized in a specific field, is initially affected to an agent. Depending on the event context, robots can automatically select the trained and appropriate trained network to resolve the situation either by using their own training models, or by collaborating with other agents specialized to perform the context event. In this article, we present results of a model implementation based on DeepLearning4J Framework and a multi-agent system middleware