Short description: A new codebook design for Vector Symbolic Architecture (VSA) aiming to mitigate the quadratic time complexity during the clean-up process that may potentially useful in VSA formed memory used in neural networks.
Short description: A new codebook design for Vector Symbolic Architecture (VSA) aiming to mitigate the quadratic time complexity during the clean-up process that may potentially useful in VSA formed memory used in neural networks.
Short description: A study in scalable external memories for neural networks’s limitations on their capacity and computational complexity. This study examined 2 directions: proving theoretical limitations on single-pass full-capacity learning rules for linear threashold model as memory, and developing a more efficient vector-symbolic key-value memory architecture with linearithmic clean-up complexity.