- Title: Towards generative models of human behaviour
- Speaker: Tim Pearce (Microsoft Research)
- Time and date: 1pm to 2pm, July 19th, 2023 (Wednesday)
- Room: Virtual (Zoom)
The Game AI Research Group is glad to announce a talk by Tim Pearce on Wednesday July 19th 2023 at 13:00. Members of DERI will be able to join us in PB.7 (basement), although space is limited.
Abstract
Large generative AI models trained across massive diverse human datasets have had huge impact on the text and image domains. This talk considers how similar techniques can be applied to sequential decision making domains, as generative models of behaviour. Specifically, I’ll present recent work using diffusion models to imitate human demonstrations (Imitating Human Behaviour with Diffusion Models, ICLR 2023), showcased in video games and robotics. I’ll also reflect on the current status of reinforcement learning in the era of foundation models.
Biography
Tim is an AI researcher at Microsoft Research in the ‘Reinforcement Learning for Games’ team. His current focus is on building agents that behave in humanlike ways, and can learn general ‘world models’ of their environments. He also has deep expertise in uncertainty estimation for neural networks. Tim completed his PhD at the University of Cambridge, spending time at the Alan Turing Institute, and did a postdoc at Tsinghua University. Tim adopts a probabilistic view of learning to help understand current algorithms and design those of the future. He believes that studying reinforcement learning settings, with their mix of sequential decisions and interactive learning, is key to making long-term progress in AI.