HTM_theory

  1. Overview
    cellular structure are same.(bio pic)

hierarchical structure.(need a pic)and each region is performing the same set of processes on the input data.

[SDR] sparse distributed representations(0 or 1)

Input: 1. Motor commands 2. sensory input

Encoder: takes a datatype and converts it into a sparse distributed representations.

Temporal means that systems learn continuously, every time it receives input it is attemptiing to predict what is going to happen next.

  1. Sparse Distributed Representation(SDR)
    Terms: 1. n=Bit array length 2. w=Bits of array 3. sparsiity 4. Dense Bit Array Capacity=2**of bits
    Bit array

Capacity=n!/w!(n-w)!

OVERLAP/UNION
Similarity can be represented by overlap/union? of SDR.

MATCH

  1. Overlap Sets and Subsampling
  2. Scalar Encoder(retina/cochlea)

    • Scalar Encoder: consecutive one
    • Random Distributed Scalar Encoder: random one
  3. Data-time encoder

  4. Input Space& Connections
    • Spactial Pooler: maintain a fixed sparsity & maintain a overlap properties.
请zzy824喝杯咖啡
0%