Scalable knowledge base completion with superposition memories
This addresses scalable knowledge graph completion for AI systems, though it appears incremental as it builds on existing neural architectures.
The authors tackled knowledge base completion by introducing Harmonic Memory Networks (HMem), which model entities as weighted sums of neighbor-relation bindings, enabling on-the-fly representation of unseen entities and achieving state-of-the-art results on benchmarks.
We present Harmonic Memory Networks (HMem), a neural architecture for knowledge base completion that models entities as weighted sums of pairwise bindings between an entity's neighbors and corresponding relations. Since entities are modeled as aggregated neighborhoods, representations of unseen entities can be generated on the fly. We demonstrate this with two new datasets: WNGen and FBGen. Experiments show that the model is SOTA on benchmarks, and flexible enough to evolve without retraining as the knowledge graph grows.