New framework improves accuracy of hyperdimensional computing codebooks

Researchers have developed a new framework to improve the accuracy of dense binary hypervector codebooks used in hyperdimensional computing and vector symbolic architectures. This framework, called transition-based derandomization, addresses inaccuracies that arise from finite dimensionality in randomized constructions. It separates the target similarity law from the derandomization variant and generator construction, allowing for explicit control over initialization, selection, update, and capacity-handling mechanisms to shape the induced similarity profile. Simulations validate the theory, offering practical guidance for selecting scalar codebook generators that better align with desired similarity laws under hardware constraints. AI

IMPACT This research could lead to more efficient and accurate hyperdimensional computing systems, potentially impacting areas like pattern recognition and data analysis.

RANK_REASON The cluster contains an academic paper published on arXiv detailing a new theoretical framework and simulation results for improving a specific area of computing.

Read on arXiv cs.NE (Neural & Evolutionary) →

• alphaXiv
• arXiv
• CatalyzeX Code Finder for Papers
• Connected Papers
• DagsHub
• Dmitri Rachkovskij A.
• Gotit.pub
• Hugging Face
• hyperdimensional computing
• information theory
• Litmaps
• ScienceCast
• scite Smart Citations

• paper
• infra

AI-generated summary · Google Gemini · from 2 sources. How we write summaries →