© 2008 A.V. Irinev
Supervisor: À.À. Shalyto
Saint-Petersburg State University of Information Technologies, Mechanics and Optics
The aim of the work is showing the new approach for construction of the operating probabalistic automatas. This approach is based on algorithms of reinforcement learning and allows to solve optimization problems for the systems having stochastic nature. In this case use of traditional training methods appears inefficient. The given approach does not work directly with probabalistic model. Instead probabalistic automata is generated on the last step of training phase.