BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Talks.cam//talks.cam.ac.uk//
X-WR-CALNAME:Talks.cam
BEGIN:VEVENT
SUMMARY:Policy transfer ensures fast learning for continuous-time LQR with
  entropy regularization - Xin Guo (University of California\, Berkeley)
DTSTART:20251111T090000Z
DTEND:20251111T094000Z
UID:TALK238438@talks.cam.ac.uk
DESCRIPTION:Reinforcement Learning (RL) enables agents to learn optimal de
 cision-making strategies through interaction with an environment\, yet tra
 ining from scratch on complex tasks can be highly inefficient. Transfer le
 arning (TL)\, widely successful in large language models (LLMs)\, offers a
  promising direction for enhancing RL efficiency by leveraging pre-trained
  models. This paper investigates policy transfer&mdash\;a TL approach that
  initializes learning in a target RL task using a policy from a related so
 urce task&mdash\;in the context of continuous-time linear quadratic regula
 tors (LQRs) with entropy regularization. We provide the first theoretical 
 proof of policy transfer for continuous-time RL\, proving that a policy op
 timal for one LQR serves as a near-optimal initialization for closely rela
 ted LQRs\, while preserving the original algorithm&rsquo\;s convergence ra
 te. Furthermore\, we introduce a policy learning algorithm for continuous-
 time LQRs that achieves global linear and local super-linear convergence. 
 Our results demonstrate both theoretical guarantees and algorithmic benefi
 ts of transfer learning in continuoustime RL\, addressing a gap in existin
 g literature and extending prior work from discrete to continuous settings
 .
LOCATION:Seminar Room 1\, Newton Institute
END:VEVENT
END:VCALENDAR