Dynamic programming: deterministic and stochastic models
Dynamic programming: deterministic and stochastic models
Artificial Intelligence - Chips challenging champions: games, computers and Artificial Intelligence
Programming backgammon using self-teaching neural nets
Artificial Intelligence - Chips challenging champions: games, computers and Artificial Intelligence
World-championship-caliber Scrabble
Artificial Intelligence - Chips challenging champions: games, computers and Artificial Intelligence
Markov Decision Processes: Discrete Stochastic Dynamic Programming
Markov Decision Processes: Discrete Stochastic Dynamic Programming
Introduction to Reinforcement Learning
Introduction to Reinforcement Learning
IJCAI'05 Proceedings of the 19th international joint conference on Artificial intelligence
Hi-index | 0.00 |
This paper deals with a nondeterministic game based on die rolls and on the "stop or continue" principle: Pickomino. During his turn, each participant has to make the best decisions first to choose the dice to keep, then to choose between continuing or stopping depending on the previous rolls and on the available resources. Markov Decision Processes (MDPs) offer the formal framework to model this game. The two main problems are first to determine the set of states, then to compute the transition probabilities. We provide in this paper original solutions to both problems: we provide (1) a compact representation of states and (2) a constructive method to compute the probability distributions, based on the partitioning of the space of roll results depending on a set of marked values. Finally, we show the efficiency of the proposed method through numerous experimental results: it turns out to be impressive compared to previous programs we developed.