In this paper, authors introduce Sandra, a neuro-symbolic reasoner which combines vectorial representations with deductive reasoning. Sandra creates a vector space guided by an ontology and performs reasoning over it. Sandra is based on the Description and Situation (DnS) ontology design pattern, which allows to infer all possible perspectives that can provide a plausible interpretation for a set of facts, even with incomplete information. The authors prove that their method is correct with respect to the DnS model and demonstrate that Sandra outperforms all the baselines, provides interpretability in the classification process, and allows control over the vector space.
Publication date: 2 Feb 2024
Project Page: Not given
Paper: https://arxiv.org/pdf/2402.00591
