This keynote introduces trigger graphs, a scalable symbolic reasoning technique enabling eficient (probabilistic) Datalog reasoning over large-scale graph stores. It also presents the new equivalence semantics for probabilistic logic programming that leads to improvements of up to 42% in neurosymbolic learning. It has been a common belief that symbolic reasoning does not scale. However, is this still true? In this talk, I will present trigger graphs, a symbolic reasoning technique that supports exact Datalog reasoning in the order of seconds over graph stores with billions of edges. Unlike the majority of commercial and open source reasoning engines, trigger graphs avoid redundant computation during reasoning by organizing the computation in a graph-like structure. The latter allows trigger graphs to support probabilistic reasoning that is more eficient even than approximate techniques. This year, trigger graphs became the driving force behind a new probabilistic logic programming semantics, the equivalence semantics. In the equivalence semantics, a probabilistic logic program induces a probability distribution over all possible equivalence relations between symbols, instead of a probability distribution over all possible subsets of probabilistic facts, as is standard in the relevant literature. We show that equivalence semantics overcomes the limitations in learning and inference of state-of-the-art neurosymbolic techniques for link prediction, rule mining, and symbolic grounding by up to 42%.
Declaration on Generative AI The author has not employed any Generative AI tools.