Portal de Periódicos da CAPES

Comparing and Evaluating Approaches to Probabilistic Reasoning: Theory, Implementation, and Applications

2012; Springer Science+Business Media; Linguagem: Inglês

10.1007/978-3-642-34179-3_2

1611-3349

Gabriele Kern-Isberner, Christoph Beierle, Marc Finthammer, Matthias Thimm,

AI-based Problem Solving and Planning

The handling of uncertain information is of crucial importance for the success of expert systems. This paper gives an overview on logic-based approaches to probabilistic reasoning and goes into more details about recent developments for relational, respectively first-order, probabilistic methods like Markov logic networks, and Bayesian logic programs. In particular, we feature the maximum entropy approach as a powerful and elegant method that combines convenience with respect to knowledge representation with excellent inference properties. While comparing the different approaches is a difficult task due to the variety of the available concepts and to the absence of a common interface, we address this problem from both a conceptual and practical point of view. On a conceptual layer we propose and discuss several criteria by which first-order probabilistic methods can be distinguished, and apply these criteria to a series of approaches. On the practical layer, we briefly describe some systems for probabilistic reasoning, and go into more details on the KReator system as a versatile toolbox for various approaches to first-order probabilistic relational learning, modelling, and reasoning. Moreover, we illustrate applications of probabilistic logics in various scenarios.