Hyperdimensional computing

Hyperdimensional computing (HDC) is a pattern-based learning algorithm that mimics the hyperdimensionality, randomness and robustness of the human brain and enables real-time massively parallel training.

HDC represents all data by hypervectors consisting of 10,000 dimensions. The error-resistant nature of HDC fits well with Computation in memory (CiM) using non-volatile memories. Even when errors happen in memories, highly accurate inference of AI is achieved. Large capacity memories in CiM enables massively parallel computation of hypervectors and thus achieves real-time, fast and accurate training.

HOME RESEARCH Hyperdimensional computing