Yes, we know Hinton and the rest of the deep learning field.   I think I
wrote a blog about deep learning last year.

Jeff

 

From: nupic [mailto:[email protected]] On Behalf Of Magnus
McWootton
Sent: Thursday, February 20, 2014 1:57 PM
To: [email protected]
Subject: [nupic-discuss] distributed reps

 

Thanks for the reply Jeff!

 

oh wow, i just understood what you said, so as it goes up the heirarchy, it
slowly becomes more of what was recorded each time,

so if you go enough levels, then it is very likely to be a whole recorded
state...  even if the input was incomplete, it slowly completes 

each time, as it changes its mind about it.  (from a threshold of 15 - with
20 synapses...)

 

Another thing was, I was wondering if you were aware of Geoffrey Hintons
work, the amazing thing I learnt off him, was checking

1 pixel at a location with a synapse, is not the same as doing it with a
slowly decreasing weight in all directions.

(the euclidean distance pixel matcher)

Its terrible to code, cause its a whole lot of work, but, he had some pretty
good results with improved unsupervised inference, unless

i was watching it wrong...

 

https://www.youtube.com/watch?v=HTNpI5R5lTU   here it is.

 

MRCW

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

Reply via email to