Efficient Probabilistic Logic Learning

Abstract
The interest in Probabilistic Logic Learning (PLL) or Statistical Relational Learning (SRL) lately has been increased. While many PLL systems have been developed these years we are still far from a general efficient framework. It is a challenging task to balance between expressiveness and computational efficiency. Each PLL framework does different assumptions and uses different technicks to calculate the results. A PLL framework like this is ProbLog. ProbLog semantics are relatively intuitive and have few assumptions compared with the majority of the PLL frameworks, but there are some significant speed overheads. In this talk I'll present briefly the PLL framework ProbLog and some first methods of improving its performance and robustness. The talk will include the explanation of two data structures: tries and Reduced Ordered Binary Decision Diagrams (ROBDDs). To follow the talk basic knowledge about logic programming and statistics will be helpful.
---
Theofrastos Mantadelis is a PhD student at the Computer Science Department of the Katholieke Universiteit of Leuven, Belgium, that is visiting us during the next two weeks. His main research topic is optimisations for statistical relational learning algorithms implemented over logic programming, currently focusing on improving the implementation of the statistical relational learning framework ProbLog.