Relations Are Channels: Knowledge Graph Embedding via Kraus Decompositions
A new research paper titled 'Relations Are Channels: Knowledge Graph Embedding via Kraus Decompositions' introduces a novel theoretical framework for Knowledge Graph Embedding (KGE). The study identifies three structural axioms—linearity, trace preservation, and complete positivity—that define relation operators as Kraus channels. This approach provides a principled foundation for KGE models, recovering existing operator-based models as special cases with Kraus rank one. The authors propose KrausKGE, a new model that effectively handles complex one-to-N and N-to-N relations, supports k-hop reasoning without explicit path encoders, and removes the need for norm constraints on entity embeddings. Additionally, the framework establishes the first theoretically grounded per-relation complexity measure in KGE literature. Empirical evaluations demonstrate that KrausKGE consistently outperforms strong baselines, particularly in scenarios with high relation fan-out, aligning with theoretical predictions. This work represents a significant advancement in machine learning and artificial intelligence, offering improved accuracy and theoretical rigor for knowledge graph representations.
Wire timeline
Relations Are Channels: Knowledge Graph Embedding via Kraus Decompositions
A new research paper titled 'Relations Are Channels: Knowledge Graph Embedding via Kraus Decompositions' introduces a novel theoretical framework for Knowledge Graph Embedding (KGE). The study identifies three structural axioms—linearity, trace preservation, and complete positivity—that define relation operators as Kraus channels. This approach provides a principled foundation for KGE models, recovering existing operator-based models as special cases with Kraus rank one. The authors propose KrausKGE, a new model that effectively handles complex one-to-N and N-to-N relations, supports k-hop reasoning without explicit path encoders, and removes the need for norm constraints on entity embeddings. Additionally, the framework establishes the first theoretically grounded per-relation complexity measure in KGE literature. Empirical evaluations demonstrate that KrausKGE consistently outperforms strong baselines, particularly in scenarios with high relation fan-out, aligning with theoretical predictions. This work represents a significant advancement in machine learning and artificial intelligence, offering improved accuracy and theoretical rigor for knowledge graph representations.
cs.AI updates on arXiv.org