Hyperdimensional computing for structured querying on tabular data embeddings

The paper presents a novel application of HyperDimensional Computing (HDC) using Holographic Reduced Representations (HRR) for structured querying on tabular data embeddings. It addresses the limitations of existing embedding methods by providing interpretable similarity scores and principled thresholds for retrieval, which are crucial for tasks like zero-match detection. The evaluation shows that HDC outperforms a graph-based baseline (EmbDI) in row retrieval across various configurations, particularly excelling in handling non-equality predicates and achieving high accuracy in attribute projection.