Paul Müller, currently working at Field AI (formerly of Deepmind), visited us on Wednesday, 6/11.
Online participants, including Matsuo Lab researchers, assigned students, and lecture students, attended the lecture and learned about multi-agent reinforcement learning methods and their applications.
Speaker biography:Paul Müller is a Multiagent Reinforcement Learning researcher. After finishing his PhD and Research Scientist position at Google DeepMind where he worked on many projects including Stratego, MuJoCo soccer and finetuning Gemini, Paul moved on to training foundational models for agentic systems at H Company, to currently working on finetuning diffusion models at Moonvalley. He is particularly interested in the application of multiagent reinforcement learning techniques on real systems, be it for reliability guarantees, or performance.
Title: Multiagent reinforcement learning and some applications to e.g. LLM training.
Abstract: In this talk, I will go over some multiagent reinforcement learning techniques and their uses, and how to use them to improve foundational models and solve different applied problems – Stratego and other 2-player 0-sum board games, LLM finetuning to maximize reliability, opponent adaptation…
Paul Müller, thank you very much for visiting Matsuo Lab.
