General methods for Search and reinforcement Learning
People
(Responsible)
External participants
Chernov Alexey
(Third-party beneficiary)
Abstract
We focus on computable, practically feasible optimal search and learning algorithms based on concepts of algorithmic information theory, using principles of Levin's nonincremental universal search and Schmidhuber's incremental, optimal ordered problem solver (OOPS) for problems of optimization and prediction. OOPS is a novel, optimally fast, incremental learner that is able to improve itself through experience. It already has been shown to solve tasks on which traditional search and learning algorithms fail. It is now being applied to to problems of game-playing and robotics.
Additional information
Start date
01.04.2002
End date
01.04.2003
Duration
12 Months
Funding sources
SNSF
Status
Ended
Category
Swiss National Science Foundation /
Project Funding /
Mathematics, Natural and Engineering Sciences (Division II)