Program for Prediction of Partially Observable Process Dynamics

The task of prediction of dynamics of partially observable processes is a very important task of artificial intelligence, because it allows us to solve a huge number of practical problems. In the diploma the modern methods of dynamics prediction are investigated, a modification for one of the methods is proposed. During the diploma a graphical application was implemented, which can be used by users to extract information about the hidden dynamics of the process from the observed data, and also to interact with the process to maximize total reward received from the process. The program also gives user an opportunity to generate video-report and pdf-report about the progress of training process. The software technologies that were used during the developing process are also described in the paper. The paper contains 34 pages, 3 chapters, 11 illustrations, 11 bibliography items, 4 appendices. Keywords: state space models, reinforcement learning, machine learning, particle filter, auto-encoding sequential monte-carlo.