Cortical Column

I am enjoying the new book A Thousand Brains. On page 65 we see A model of a cortical column. I have a few questions. How does it learn? What s the learning rule(s). How does the locations level select what basis vectors to use?

How does the observed features level distinguish between feature as observed versus feature as evoked by the locations level. Evolution needs to clearly differentiate between observed and imagined. It lowers your survivability if you are psychotic.

Does Numenta have a software emulation of a cortical column learning?

How can we best implement a cortical column in hardware?

Are white matter connections DNA determined or learned?

I very much agree the smart cortical column is a big step forward. I am looking forward to seeing it used to advance AI.

they have still only partial understanding … so far …

the L4 and L6 are sort of sequence learners/predictors, where L6 is also grid-like location mapper.
L4 <–> L6 are connected in an interloop to “narrow” the Location-Landmark/feature recognition in a manner similar to kalman filter/gradient-descent …

Once CC recognizes the a Landmark/feature other CC can use it in “frame-set” to build more complex structures.

Say CC1 recognizes cup handle, CC2 recognizes the cup-bulk … CC1-CC2 recognizes a cup … CCx uses the env-frame to plan the putting the cup on the table… something like that … still assimilating the book

1 Like