A Novel Framework for Multiagent Knowledge-Based Federated Learning Systems.

Multiagent systems promote a decentralized and distributed approach that enable the division of complex problems into smaller parts. The use of multiagent systems also enables the representation of physical entities, such as persons, pursuing their own goals in an active and proactive society. Currently developments are promoting the idea of having machine learning models in agents to enable intelligent decisions in agents-side. However, machine learning, required assess to large datasets that cannot be available locally to individual agents, demanding the sharing of data or the use of public available datasets to training models for a given agent. To address this issue, this paper proposes the use of federated learning to enable the existence of a collaborative learning model that respects the data privacy, security, and ownership and can be in compliance with the European General Data Protection Regulation (EU GDPR). This paper proposes a novel framework called Python-based framework for agent-based communities powered by federated learning (PEAK FL) that will provide all the necessary tools to build powerful federated learning solutions based on agent communities. This framework provides the users the ability to implement and test hybrid solutions (multiagent-based federated learning systems) in a simple-to-use way, removing the unnecessary boilerplates.