Gestion du raisonnement à base de cas avec l'apprentissage par renforcement pour un jeu contraint dans le temps
|Advisor:||Lamontagne, Luc D.|
|Abstract:||In this work, we try to improve the behavioral aspects of video games using Case Based Reasoning (CBR), which can reproduce human behavior as reasoning by similarity, as well as remembering and forgetting previous experiences. This technique, coming from the Artificial Intelligence field, solves new problems by retrieving similar past experiences in the case base and adapting solution to solve new problems. We use CBR for the automation of decisions made by the game engine. The construction of a CBR system needs to accumulate many episodes from the gaming environment to create the case base of the CBR engine. However, as the number of episodes being saved in the case base increases, the response time of the CBR system slows down. We are then facing a dilemma: reducing the size of the case base to improve the response of the CBR system while keeping an acceptable level of performance. In this master thesis, we use the game of Tetris to conduct our case studies. This game presents some particular interests, as decisions to be made are limited by time constraints. We propose in this thesis to answer the following questions: How to construct a CBR system to play the game of Tetris. What is the expected performance of the system applied to this game? Wich game level can be reached by estimating case value through reinforcement learning? As time response constraints are inherent to Tetris, which degradation of performance can be expected by removing cases from the case base?|
|Document Type:||Mémoire de maîtrise|
|Open Access Date:||16 April 2018|
|Collection:||Thèses et mémoires|
All documents in CorpusUL are protected by Copyright Act of Canada.