First-Order Stochastic Action Language (2002)
| Venue: | Artificial Intelligence |
BibTeX
@ARTICLE{Watanabe02first-orderstochastic,
author = {Hiroaki Watanabe and Stephen Muggleton},
title = {First-Order Stochastic Action Language},
journal = {Artificial Intelligence},
year = {2002},
volume = {7}
}
Bookmark
 |
 |
 |
 |
 |
 |
OpenURL
Abstract
Action languages are high-level languages for describing changing worlds. We propose two new stochastic action languages for increasing robustness under uncertainties. We introduce a stochastic version of the action language A at propositional level and upgrade it to Datalog Stochastic Action Language (DatalogSAL) in order to handle dynamic worlds at Datalog level. A model of Datalog-SAL is de ned at dynamic Bayesian networks (DBNs) level. Datalog-SAL assigns probability distributions over state transitions, actions, and initial states. Datalog-SAL provides a simple and smooth conversion of Datalog and probability in the action and change framework. Since a model of Datalog-SAL is given in DBNs, learning Datalog-SAL would be possible via DBNs' parameter learning algorithms. A complexity analysis of learning Datalog-SAL is discussed in terms of the Baum-Welch parameter learning algorithm.
