Scallop: A Language for Neurosymbolic Programming

We present Scallop, a language which combines the benefits of deep learning and logical reasoning. Scallop enables users to write wide range neurosymbolic applications train them in data- compute-efficient manner. It achieves these goals through three key features: 1) flexible symbolic representation that is based on relational data model; 2) declarative logic programming Datalog supports recursion, aggregation, negation; 3) framework for automatic efficient differentiable reasoning theory provenance semirings. evaluate suite eight from literature. Our evaluation demonstrates capable expressing algorithmic diverse challenging AI tasks, provides succinct interface machine programmers integrate domain knowledge, yields solutions are comparable or superior state-of-the-art models terms accuracy. Furthermore, Scallop's outperform aspects such as runtime efficiency, interpretability, generalizability.