Ying Wen, University College London
May 28, 2018, Mon, 15:00-16:00
Most successful researches on reinforcement learning have been in single agent domain. However, many complex reinforcement learning problems such as multiplayer games, machine bidding in competitive e-commerce and financial markets are naturally modelled as multi-agent systems. Moreover, the interactions between agents are introduced on the basis of interactions with the environment, therefore resulting in unstable and more challenging training and optimisation in multi-agent reinforcement learning (MARL).
In this talk, I shall explicit the theoretical Stochastic Game setup for MARL and give an overview of the many challenges specific to the multi-agent aspect. Within this perspective, the notion of communication inherent to the multi-agent system and its different elements would be explored. I shall also give some actual applications cases and methods to address the communication in MARL.
Ying Wen is a PhD Candidate in the Department of Computer Science, University College London. His research interests include reinforcement learning and deep learning techniques for real-world scenarios, such as computational advertising, multi-agent system. He has published several papers in international journals and conferences, such as AAMAS, IJCAI, ICDM. Ying earned his MRes with Distinction Honor from University College London in 2016 and B.Eng. with First Class Honor from Queen Mary, University of London and Beijing University of Posts and Tel. in 2015. He was an intern at MediaGamma, Amazon and Baidu.