Paper available

[Szepesvari Csaba]

The following paper is available from

http://victoria.mindmaker.hu/~szepes/papers/scann98.ps.gz

Reinforcement Learning: Theory and Practice 

Cs. Szepesvri 

in Proceedings of the 2nd Slovak Conference on Artificial Neural
Networks (SCANN'98).

Nov. 10-12, 1998, Smolenice, Slovakia, pp. 29-39 (Ed: Marian Hrehus)

   We consider reinforcement learning methods for the solution of
complex sequential optimization problems. In particular, the soundness of
two methods proposed for the solution of partially observable problems
will be shown.
  
The first method is a state-estimation scheme and requires mild {\em
a priori} knowledge, while the second method assumes that a significant
amount of abstract knowledge is available about the decision problem and
uses this knowledge to setup a macro-hierarchy to turn the partially
observable problem into another one which can already be handled using
methods worked out for observable problems. This second method is also
illustrated with some experiments on a real-robot.

--------------------------------------------------------------------
Csaba Szepesvari

Head of Research Department

Mindmaker Ltd.
Budapest 1112
Konkoly-Thege Miklos u. 29-33.
HUNGARY

e-mail: szepes at mindmaker.hu
WEB:    http://victoria.mindmaker.hu/~szepes
Phone:  +36 1 395 9220/1205 (dial extension continuously)
Fax:    +36 1 395 9218