Project: Nature-Inspired Communication in Reinforcement Learning Systems

Key information:

Abstract: 

From the beginning, artificial intelligence has innovated by borrowing concepts from nature. Previous work has shown that certain communication networks are beneficial for systems with multiple reinforcement learning agents, but more work needs to be done to analyse these communication networks and optimise them with machine learning. We aim to use well-established methods for understanding networks from ecology, sociology and mathematics. With these methods, our research will have two complementary goals: Firstly, to improve the analysis of communication networks in distributed and multi-agent reinforcement learning systems. Secondly, to provide further insights into the parallels between reinforcement learning agents, humans and animals. We believe that both goals will lead to new opportunities for technological innovation, particularly for robotics and the distributed control of IoT systems.