Hyperdimensional Computing
Overview
Hyperdimensional computing represents data with long vectors whose distributed patterns can support learning and memory even when individual components are noisy or quantized. Methods such as binary HDC, Holographic Reduced Representations (HRR), and Fourier HRR (FHRR) encode features and sequences using operations like bundling, binding, and permutation, then classify or retrieve information with similarity search and associative memory. These approaches are studied for tasks including time-series and image classification, anomaly detection, structured data retrieval, and edge or federated learning, with hardware designs targeting low-power and in-memory computation. A central tension is that compact, robust representations can be efficient to store and process, yet superposition can make precise retrieval and multi-step compositional reasoning difficult.
Latest papers
39 papers5 papers in the last four weeks, against 2 the four weeks before. 0.0% of all new papers.