BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Talks.cam//talks.cam.ac.uk//
X-WR-CALNAME:Talks.cam
BEGIN:VEVENT
SUMMARY:Breaking the Sample Size Barrier in Reinforcement Learning  - Yuxi
 n Chen\, University of Pennsylvania
DTSTART:20251112T140000Z
DTEND:20251112T150000Z
UID:TALK234916@talks.cam.ac.uk
CONTACT:Dr Varun Jog
DESCRIPTION:Abstract: Emerging reinforcement learning (RL) applications ne
 cessitate the design of sample-efficient solutions in order to accommodate
  the explosive growth of problem dimensionality. Despite the empirical suc
 cess\, however\, our understanding about the statistical limits of RL rema
 ins highly incomplete. In this talk\, I will present some recent progress 
 towards settling the sample complexity limits in RL. The first scenario is
  concerned with RL with a generative model\, which allows one to query arb
 itrary state-action pairs to draw independent samples. We prove that a mod
 el-based algorithm (a.k.a. the plug-in approach) achieves minimal-optimal 
 sample complexity without any burn-in cost. The second scenario is concern
 ed with online RL\, where an agent learns via real-time interactions with 
 an unknown environment. We develop the first algorithm — an optimistic m
 odel-based algorithm — that achieves minimax-optimal regret for the enti
 re range of sample sizes. Time permitting\, we will also discuss the effec
 tiveness of model-based paradigms in offline RL and multi-agent RL. Our re
 sults emphasize the prolific interplay between high-dimensional statistics
 \, online learning\, and game theory.\n\nThe first part is based on joint 
 work with Gen Li\, Yuting Wei and Yuejie Chi\, and the second part is base
 d on joint work with Zihan Zhang\, Jason Lee and Simon Du.\n\nPaper 1: htt
 ps://arxiv.org/abs/2005.12900\nPaper 2: https://yuxinchen2020.github.io/pu
 blications/Optimal-OnlineRL.pdf\nPaper 3: https://arxiv.org/abs/2204.05275
 \n\nBio: Yuxin Chen is currently a professor of statistics and data scienc
 e and of electrical and systems engineering at the University of Pennsylva
 nia. Before joining UPenn\, he was an assistant professor of electrical an
 d computer engineering at Princeton University. He completed his Ph.D. in 
 Electrical Engineering at Stanford University and was also a postdoc schol
 ar at Stanford Statistics. His current research interests include high-dim
 ensional statistics\, machine learning theory\, and optimization. He has r
 eceived the Alfred P. Sloan Research Fellowship\, the SIAM Activity Group 
 on Imaging Science Best Paper Prize\, the ICCM Best Paper Award (gold meda
 l)\, and was selected as a finalist for the Best Paper Prize for Young Res
 earchers in Continuous Optimization. He has also received the Princeton Gr
 aduate Mentoring Award.\n\n
LOCATION:MR5\, CMS Pavilion A
END:VEVENT
END:VCALENDAR