Utilizing DeepDive/Tuffy for more scalable statistical inference? about factorbase

[vidhiJain]

• Tuffy utilizes WalkSAT, a stochastic local search algorithm that is same as in Alchemy. It uses heuristic based partitioning to introduce parallelism for speed up. The paper states

In contrast to probability models like factor graphs [23] that require complex distributions to be specified in tedious detail, MLNs allow us to declare a rigorous statistical model at a much higher conceptual level using first-order logic.

How far this statement holds true?

• DeepDive seems a significant related work. It conducts statistical inference on factor graphs to determine marginal probability by Gibbs sampling. It uses DimmWitted inference engine.