Differentiable Reasoning about Knowledge Graphs with Reshuffled Embeddings
Keywords: Differentiable reasoning, knowledge graphs, region based embeddings
TL;DR: We propose a new region based knowledge graph embedding model and show that is capable of faithfully capturing different classes of rule bases.
Abstract: Knowledge graph (KG) embedding methods learn geometric representations of entities and relations to predict plausible missing knowledge. These representations are typically assumed to capture rule-like inference patterns. However, our theoretical understanding of the kinds of inference patterns that can be captured in this way remains limited. Ideally, KG embedding methods should be expressive enough such that for any set of rules, there exists an embedding that exactly captures these rules. This principle has been studied within the framework of region-based embeddings, but existing models are severely limited in the kinds of rule bases that can be captured. We argue that this stems from the use of representations that correspond to the Cartesian product of two-dimensional regions.
As an alternative, we propose RESHUFFLE, a simple model based on ordering constraints that can faithfully capture a much larger class of rule bases than existing approaches. Moreover, the embeddings in our framework can be learned by a Graph Neural Network (GNN), which effectively acts as a differentiable rule base. This has some practical advantages, e.g. ensuring that embeddings can be easily updated as new knowledge is added to the KG. At the same time, since the resulting representations can be used similarly to standard KG embeddings, our approach is significantly more efficient than existing approaches to differentiable reasoning. The GNN-based formulation also allows us to study how bounded inference can be captured. We show in particular that bounded reasoning with arbitrary sets of closed path rules can be captured in this way.
Primary Area: neurosymbolic & hybrid AI systems (physics-informed, logic & formal reasoning, etc.)
Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics.
Submission Guidelines: I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2025/AuthorGuide.
Anonymous Url: I certify that there is no URL (e.g., github page) that could be used to find authors’ identity.
No Acknowledgement Section: I certify that there is no acknowledgement section in this submission for double blind review.
Submission Number: 3106
Loading