sLTN: Extending Logic Tensor Networks with Structural Dimensions
A recent submission to arXiv (2608.11136) presents sLTN (Structural Logic Tensor Networks), which expands upon Logic Tensor Networks (LTN) by integrating structural aspects such as temporal sequence and graph connectivity into the neurosymbolic paradigm. The authors define the syntax and fuzzy tensor semantics of sLTN, facilitating temporal, sequential, and relational reasoning within differentiable learning. This advancement addresses a shortcoming of LTN, which fails to account for structural organization. In sLTN, structural dimensions are represented as named tensor axes for both quantification and relation. This research holds significance for artificial intelligence and machine learning, especially within neurosymbolic AI, impacting applications involving structured data.
Key facts
- sLTN is an extension of Logic Tensor Networks (LTN) that incorporates structural dimensions.
- Structural dimensions represent named tensor axes for temporal order, sequence position, or graph nodes.
- sLTN allows explicit quantification and structural relations at the logical level.
- The paper formalizes the syntax and fuzzy tensor semantics of sLTN.
- The original LTN is suited for flat collections of individuals and lacks structural organization.
- The paper is available on arXiv with identifier 2608.11136.
- The announcement type is 'new'.
- The full abstract is truncated in the provided content.
Entities
Institutions
- arXiv