BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Talks.cam//talks.cam.ac.uk//
X-WR-CALNAME:Talks.cam
BEGIN:VEVENT
SUMMARY:Adaptive Partitioning and Learning for Stochastic Control of Diffu
 sion Processes - Hanqing Jin (University of Oxford)
DTSTART:20251111T113000Z
DTEND:20251111T121000Z
UID:TALK238465@talks.cam.ac.uk
DESCRIPTION:We study reinforcement learning for controlled diffusion proce
 sses with unbounded continuous state spaces\, bounded continuous actions\,
  and polynomially growing rewards&mdash\;settings that arise naturally in 
 finance\, economics\, and operations research. To overcome the challenges 
 of continuous and high-dimensional domains\, we introduce a model-based al
 gorithm that adaptively partitions the joint state&ndash\;action space. Th
 e algorithm maintains estimators of drift\, volatility\, and rewards withi
 n each partition\, refining the discretisation whenever estimation bias ex
 ceeds statistical confidence. This adaptive scheme balances exploration an
 d approximation\, enabling efficient learning in unbounded domains. Our an
 alysis establishes regret bounds that depend on the problem horizon\, stat
 e dimension\, reward growth order\, and a newly defined notion of zooming 
 dimension tailored to unbounded diffusion processes. The bounds recover ex
 isting results for bounded settings as a special case\, while extending th
 eoretical guarantees to a broader class of diffusion-type problems.&nbsp\;
LOCATION:Seminar Room 1\, Newton Institute
END:VEVENT
END:VCALENDAR