Hi,
That would be awesome. Thanks Jeff Hawkins.

Matt Taylor : 
That is so cool. I would love to read about every new temporal pooler 
hypothesis.
I have a working prototype of CLA that I built recently. Its exactly what CLA 
is in white papers and nupic videos. Implementation is little different from 
Nupic. eg: spatial pooler is called connection_matrix in that prototype. Its a 
matrix with random connections. Connections in between cells are stored in a 
matrix and learning literally strengthens those connections. 
SoN idea i proposed, I am planning on trying it out with this CLA prototype. As 
far as I have hypothesized, it does form a higher level representation. one 
whole sentence is represented by few cells in SDR[3]. Plus the integral motor 
implementation also is one of my plan. SO instead of having different CLAs, the 
same ones would talk back.
I would be able to demonstrate higher level representation of a complete 
sentence, with different representations for every word. eg: HOW ARE YOU will 
have a different pattern then ARE HOW YOU. sequence is important. Plus I am 
excited to have completely understood the idea of semantic similarity with 
shared bits. This prototype will have some of the bits of "HOW ARE YOU" if the 
sentence it gets is "WHAT ARE YOU"  as both share "ARE YOU" and that too in 
same sequence.

That SoN idea I wrote was hard for you to follow because I was incapable of 
putting it in a simple way. Its very simple and intuitive. Once new version is 
ready I will update my post.

Regards
Aseem Hegshetye

_______________________________________________
nupic mailing list
[email protected]
http://lists.numenta.org/mailman/listinfo/nupic_lists.numenta.org

Reply via email to