文摘
We offer a new instantiation perspective by encoding the ground substitutions as simple paths in the Herbrand universe. We scale up MLN learning by combining the benefits of random walks and subgraph pattern mining, which avoids exploring the entire POG. We ensure efficient MLN inference by constructing the template network to locate promising paths that can ground the given clauses. We provide the computational complexity analysis, demonstrating that the time complexity of our framework is independent of the size of the KB.