Conjunctive block coding for hyperdimensional graph representation

Ali Zakeri, Zhuowen Zou, Hanning Chen, Hugo Latapie, Mohsen Imani

Abstract: Highlights•Knowledge Graph Enhancement: CLOG presents a novel hyperdimensional graph representation for improved graph task performance.•Increased Memory Capacity: Leveraging HDC's variable binding, CLOG boosts memory capacity for sparse graphs.•Efficient Decoding Scheme: A block encoding and masking scheme ensures efficient decoding for hyperspace projections.•Integration with GNN: CLOG seamlessly integrates with GNN, showcasing its versatility in structural representation.•Benchmark Performance: CLOG shows competitive performance on link prediction datasets compared to existing methods.

External IDs:dblp:journals/iswa/ZakeriZCLI24