Inference in Probabilistic Answer Set Programming Under the Credal Semantics

Probabilistic Answer Set Programming under the credal semantics (PASP) describes an uncertain domain through an answer set program extended with probabilistic facts. The PASTA language leverages PASP to express statistical statements. A solver with the same name allows to perform inference in PASTA programs and, in general, in PASP. In this paper, we investigate inference in PASP, propose a new inference algorithm called aspcs based on Second Level Algebraic Model Counting (2AMC), and implement it into the aspmc solver. Then, we compare it with PASTA on a set of benchmarks: the empirical results show that, when the program does not contain aggregates, the new algorithm outperforms PASTA. However, when we consider PASTA statements and aggregates, we need to replace aggregates with a possibly exponential number of rules, and aspcs is slower than PASTA.