Stochastic Relational Processes and Models: Learning and Reasoning (Stochastisch relationele processen en modellen: leren en redeneren)

In order to solve real-world tasks, intelligent machines need to be able to act in noisy worlds where the number of objects and the number of relations among the objects varies from domain to domain. Algorithms that address this setting fall into the subfield of artificial intelligence known as statistical relational artificial intelligence (StaR-AI).While early artificial intelligence systems allowed for expressive relational representations and logical reasoning, they were unable to deal with uncertainty. On the other hand, traditional probabilistic reasoning and machine learning systems can capture the inherent uncertainty in the world, but employ a purely propositional representation and are unable to capture the rich, structured nature of many real-world domains.StaR-AI encompasses many strains of research within artificial intelligence. One such direction is statistical relational learning which wants to unify relational and statistical learning techniques. However, only a few of these techniques support decision making processes.This thesis advances the state-of-the-art in statistical relational learning by making three important contributions. The first contribution is the introduction of a novel representation, called causal probabilistic time-logic (CPT-L) for stochastic relational processes. These are stochastic processes defined over relational state- spaces and they occupy an intermediate position in the expressiveness/efficiency trade-off. By focusing on the sequential aspect and deliberately avoiding the complications that arise when dealing with hidden states, the algorithms for inference and learning for CPT-L are more efficient than those of general purpose statistical relational learning approaches. The second contribution is that we show how to adapt and generalize the algorithms developed for CPT-L so that they can be used to perform parameter estimation in the probabilistic logic programming language ProbLog. The final contribution of this thesis is a decision theoretic extension of the ProbLog language that allows to represent and to solve decision problems.status: publishe