An algebraic ProbLog and its applications.

Reasoning about the environment is central to many tasks in artificial intelligence. While traditional approaches based on first order logic are well suited to reason about a single, fixed situation, in practice one often has to compare or combine multiple options based on some property. For instance, the cost or probability of achieving a given goal can be estimated based on costs or probabilities of subgoals. This can be seen as a labeling task, where labels are given to basic statements in a logical theory and the task is to compute labels for derived statements, called queries. Previously, we have introduced ProbLog, a popular probabilistic logic programming language where probability labels define a distribution over possible worlds, which is used to calculate probabilities of queries. However, labeling tasks in other settings require both different types of labels and more flexible ways to combine these. The goal of this project therefore is to develop a language for modeling such tasks as well as corresponding reasoning mechanisms, based on the following central questions: - What are the general characteristics of such labeling tasks? - How can efficient inference be realized? - How can (parts of) models be learned automatically?

Keywords:Statistical relational learning, Possible worlds semantics, Probabilistic programming, Logic programmin

Disciplines:Applied mathematics in specific fields