Abstract
Multi-Agent Reinforcement Learning (MARL) provides a framework for modelling complex systems in which multiple decision-makers interact within a shared environment. Such settings are ubiquitous in the real world, yet introduce fundamental challenges that do not arise in single-agent learning. This talk presents an accessible overview of MARL, starting from the core principles of reinforcement learning and extending to multi-agent formulations grounded in game theory and decentralised decision-making. It highlights key difficulties that emerge at scale such as coordination, non-stationarity, and partial observability and surveys the main approaches used to address them. Finally, the talk offers a perspective on recent advances in the field, with a focus on developing scalable and efficient algorithms for real-world multi-agent systems.
Bio
Ruan de Kock is a Research Engineer at InstaDeep and an MSc student at the University of Cape Town, supervised by Jonathan Shock. He specialises in multi-agent reinforcement learning, with research focused on developing advanced algorithms for solving complex problems at scale. His work has been published at leading venues, including NeurIPS, ICML, ICLR, and AAMAS.