Probabilistic Logic Programming (PLP) combines two major kinds of reasoning, probabilistic and logical. PLP features drawing inference from default assumptions (closed-world semantics), and most PLP systems offer rich data structures as in traditional programming languages. This makes them well-suited for many knowledge representation applications, e.g., situational awareness under uncertainty. PLP programs consist of if-then rules that are labeled with probabilities. Learning is supported in terms of learning the probabilities of given rules, or (less common) learning the rules. More recently, research has gained momentum on combining logical with neural networks (again) and LLM reasoning, also in the context of probabilistic logic programming.

In the talk I will provide an overview of some of the above developments in context of own work. I will explain the main ideas behind our Fusemate PLP system. I will also discuss more recent developments around learning and combination with NN and LLM reasoning.

I am a Principal Research Scientist with Data61 since 2016 and a Honorary Associate Professor with the ANU CECC. Before that, I was with NICTA from 2005-2016, and before that I worked at the Max-Planck-Institut, and at several Universities in Germany (Koblenz, Giessen and Muenchen). PhD in 1996. I researched extensively on automated reasoning for classical first-order logic, in terms of proof calculi, implementing systems and applications When I joined Data61 I became interested in reasoning under uncertainty. I worked on probabilistic planning with Markov Decision Processes in combination with temporal logics. More recently I (re-)discovered my interest in logic programming and its probabilistic extension. My current research is mainly built around the development of the probabilistic logic programming system Fusemate and its application in Data61 projects.