TensorLog (Cohen et al., 2017) is an implementation of probabilistic knowledge bases that (i) leverages deep learning to answer queries efficiently, by (ii) restricting the first order logic language to a class called p-tree-DKG, and (iii) learns weights on knowledge base facts.
(Read more)Logic Tensor Networks (Serafini & Garcez, 2016) is a framework for learning and reasoning combining neural-network capabilities with knowledge base data representation structures using first-order logic.
They address the problem of (i) query answering in the unit interval [0,1] (also known as fuzzy or real logic), as well as, (ii) subsymbolic (vector) representation finding of constant symbols, function symbols and predicates.
(Read more)This paper (Rocktäschel & Riedel, 2017) presents Neural Theorem Provers (NTP), a framework for fuzzy/real logic i.e. [0,1]-query answering with backwards chaining and subsymbolic representation learning.
(Read more)