Jump to content

Marcus Hutter

From Wikipedia, the free encyclopedia

This is an old revision of this page, as edited by IDSIAupdate (talk | contribs) at 09:58, 17 March 2006 (introduction). The present address (URL) is a permanent link to this revision, which may differ significantly from the current revision.

(diff) ← Previous revision | Latest revision (diff) | Newer revision → (diff)

Marcus Hutter (* 1967) was born and educated in Munich, where he studied physics and computer science. In 2000 he joined Juergen Schmidhuber's group at the Swiss AI lab IDSIA, where he developed the first mathematical theory of optimal Universal Artificial Intelligence, based on Ray Solomonoff's theory of universal inductive inference.

Suppose you want to maximize your future expected reward in some unknown dynamic environment, up to some future horizon. This is the general reinforcement learning problem. Solomonoff's / Hutter's only assumption is that the reactions of the environment in response to your actions follow some unknown but computable probability distribution. At any time, given the limited observation sequence so far, what is the Bayes-optimal way of selecting the next action? Hutter proved that the answer is: use Solomonoff's universal prior to predict the future, and execute the first action of the action sequence that will maximize the predicted reward up to the horizon.

This is mainly a theoretical result. To overcome the problem that Solomonoff's prior is incomputable, in 2002 he also published a remarkable asymptotically fastest algorithm for all well-defined problems.