sLTN: Structural Logic Tensor Networks
sLTN adds named structural axes to Logic Tensor Networks so time, sequence position, and graph connectivity can be expressed inside the logic itself.
The paper frames standard LTN as strongest on flat collections of individuals. sLTN makes structural dimensions first-class, allowing explicit quantification over axes and relations between them. The authors formalize its syntax and fuzzy tensor semantics, and note that ordinary LTN falls out as a special case when those dimensions are absent. They also describe a PyTorch implementation tied to the companion library. ArXiv · AI/CL/LG's note
The paper frames standard LTN as strongest on flat collections of individuals. sLTN makes structural dimensions first-class, allowing explicit quantification over axes and relations between them. The authors formalize its syntax and fuzzy tensor semantics, and note that ordinary LTN falls out as a special case when those dimensions are absent. They also describe a PyTorch implementation tied to the companion library. ArXiv · AI/CL/LG's note
score 4