The purpose and how does TP work?

Science Numenta Theory
1

How does TP work. What is the goal ?

Things I understand.

  1. TP is MANY to ONE mapping
  2. The output has to be unique and stable
  3. Used as input to higher levels
  4. ??? Feedback to TM ??? does it ? how ? what does it change about TM ?

The question is how does it influences the TM predictions OR does it, should it.

The feedback is not used to help predict better the next SDR, right ? That should be the task of feedback from upper layers ?

If it does then, From the snippet below :
IF Apical & Basal segments are active This cell in the column win … I understand that.
But this does not explain how does this influences the other columns, which is the most important thing i.e. output SDR.

https://github.com/numenta/htmresearch/blob/master/htmresearch/algorithms/apical_tiebreak_temporal_memory.python2

A generalized Temporal Memory with apical dendrites that add a “tiebreak”.
Basal connections are used to implement traditional Temporal Memory.
The apical connections are used for further disambiguation. If multiple cells
in a minicolumn have active basal segments, each of those cells is predicted,
unless one of them also has an active apical segment, in which case only the
cells with active basal and apical segments are predicted.
In other words, the apical connections have no effect unless the basal input
is a union of SDRs (e.g. from bursting minicolumns).

1 Like
2

IMO, TP should be happening in another layer (i.e. not the same population of cells as TM), and it should be providing feedback to the TM layer. It should also have long distance connections with other cortical columns, allowing them to vote on the higher level concept that is being sensed. This should in turn improve the predictions made by the TM layer.

2 Likes