The HTM model Neuron and High-order sequence memory for Numeric input


#1

Hi,

How high-order memory property of htm model, does the prediction for number inputs (e.g. 20 in timestamp 0, 20 in timestamp 1, 20 in timestamp n).

I have read around this topic, but most of examples and discussion talking about the sequences of “ABCD” and “YBCX”… I work with numeric value and would like to ensure I have understood the concept.

Thanks.


#2

To understand this concept, you must have a good idea of:

  • why sparsity is important for semantic representation
  • basic number encoding techniques
  • cyclic encoding techniques
  • date-time encoding techniques

Scalar encodings can be combined with date-time encodings to add time semantics to spatial data in this way. Suggested resources:


#3

Thanks @rhyolight , one more question , I am trying to find an updated information about regions in HTM. can you please let me know where I can find this information.

I read the presentation from Jeff Hawkins is published in 2014 (Sensory-Motor Integration in HTM Theory). it is mentioned that region 2/3 is implemented , 4 is 90% understood and tested, region 5 is 50% understood and layer 6 is 10% understood.

Are these figures still valid or they are changed, any source please?

Regards.


#4

These are not regions they are called “layers” as in the horizontal layers of neocortex. Regions are larger section of cortex that include all layers within it.

There’s lots of active discussion about this all over the forum, just search.


#5

Thanks for correcting me