BoxLitE: A Faithful Knowledge Base Embedding Based on Convex Optimization
The paper introduces BoxLitE, a knowledge base embedding model that utilizes convex optimization. This model aims to enhance the representation of knowledge by mapping concepts to convex regions in a vector space. The authors demonstrate that BoxLitE can create embeddings that maintain desirable properties of faithfulness for knowledge bases.
- ▪BoxLitE is designed for DL-Lite$^{\mathcal{H}}$ knowledge bases.
- ▪The model leverages convex optimization to improve knowledge representation.
- ▪The paper presents a proof of concept for formulating the embedding task as a convex optimization problem.
Opening excerpt (first ~120 words) tap to expand
Computer Science > Artificial Intelligence arXiv:2605.23937 (cs) [Submitted on 27 Apr 2026] Title:BoxLitE: A Faithful Knowledge Base Embedding Based on Convex Optimization Authors:Bruno F. Lourenço, Hesham Morgan, Ana Ozaki, Aleksandar Pavlović, Emanuel Sallinger View a PDF of the paper titled BoxLitE: A Faithful Knowledge Base Embedding Based on Convex Optimization, by Bruno F. Louren\c{c}o and 4 other authors View PDF HTML (experimental) Abstract:Knowledge base (KB) embeddings aim at combining the capability of classical knowledge graph embeddings to generalize the information present in facts, the ABox, with conceptual knowledge represented in an ontology language, the TBox. Several authors have recently explored the idea of mapping concepts to convex regions in a vector space.
…
Excerpt limited to ~120 words for fair-use compliance. The full article is at arXiv cs.AI.
Discussion
0 commentsMore from arXiv cs.AI
