Hi Ian,

Thanks for the note. I agree with you that its important. I am not focusing on 
it at this point since it's not currently used by us here at Grok. Once I 
finish restructuring the functionality we are using, I can assess how difficult 
it would be to implement cloning/weight sharing and we can continue this 
discussion. How does that sound?

David Ragazzi – if you're following this thread – Once I am finished with the 
python implementation of the restructured spatial pooler I plan on writing a 
pure C++ version. So you might want to hang on a few weeks before writing one 
yourself!

Gil.

From: Ian Danforth <[email protected]<mailto:[email protected]>>
Reply-To: "NuPIC general mailing list." 
<[email protected]<mailto:[email protected]>>
Date: Tuesday, July 16, 2013 12:38 PM
To: "NuPIC general mailing list." 
<[email protected]<mailto:[email protected]>>
Subject: Re: [nupic-dev] Restructuring Spatial Pooler

Gil,

 A very cool project. Regarding the cloned region, I think this one is actually 
pretty important. Weight sharing is more common than not in convolutional nets, 
and having an explicit class that looks more like this strategy will make it 
easier and more accessible for other researchers to work with. If your 
implementation looks enough like weight sharing, then perhaps it should be 
called 'shared weight region' to make it obvious CLA is paying attention to and 
participating in mainstream research. If, on the other hand, it doesn't end up 
looking at all like weight sharing then Subutai is probably right in that it 
can wait.

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

Reply via email to