14:15h - 15:45h

Raum 057, Oet. 67

-------------------------------------------------------

Jan Peter: Visualization for Online Monte Carlo Planning

Bachelorarbeit, betreut von Lenz Belzner

Monte Carlo Tree Search (MCTS) is a powerful new method in AI research based on random sampling, which can be applied to various problem domains. The extensive research interest this topic has received in the past years has lead to a wide range of possible extensions and parameters which can be used to improve and refi ne the basic algorithm. Tuning the different parameters and strategies, and evaluating their effects on the final decision graph, however, can be tedious and error-prone.

In this thesis, we discuss how a toolkit can be developed that provides an interactive visualization of MCTS decision graphs for arbitrary problem domains. Besides leading to a deeper understanding of how the MCTS algorithm behaves, the toolkit allows the user to gain greater insight into the effectiveness of chosen parameters and to isolate errors in his implementation. The problem of a constantly changing and evolving planning algorithm is addressed by a highly extensible design that can be easily adapted to new approaches.