Hey, Doug, thanks for digging up the IBM papers!  Eye opening.

IBM's approach to their model, silicon, and programming environment seems quite 
rigorous and advanced.  I don't claim exhaustive knowledge of the field, but a 
lot of the prior work I've seen struck me as people randomly wiring spiking 
neurons and merely seeking signs of biologically plausible synchronous firing 
patterns, not seriously trying to construct a usable, programmable neuromorphic 
computing paradigm.  Impressive.  

Interesting to contemplate, as you suggest, how the predictive framework of the 
CLA might be expressed in "corelets"...

-Steve O.

On Aug 8, 2013, at 5:39 PM, Doug King <[email protected]> wrote:

> Hi Steve,
> 
> The last link is most interesting. I have been following the DARPA SyNAPSE 
> program output for a while but these new announcements are exciting news. 
> 
> IBM has created silicon that uses 1272 ASIC gates to create a model of a 
> leaky integrate-and-fire (LIF) neuron. The current chip has 256-neuron, 
> 64k/256k-synapse neurosynaptic cores in 45nm silicon. The leaky model with a 
> constant leak is described by five basic operations: 1. synaptic integration, 
> 2. leak integration, 3. threshold,
> 4. spike firing, and 5. reset. With that model a programmer can create higher 
> level constructs. It looks flexible enough that I could imagine using 
> something like this as an accelerator to the CLA. It's clear to me that the 
> CLA we are working with now in VonNumann software will be soon run on 
> neuromorphic co-processor hardware. This is something to keep in mind as we 
> build out the CLA and HTM codebase - how will we accelerate in hardware and 
> how can we keep a migration path open for the models we invest in without 
> loosing that investment?
> 
> Whitepapers here: 
> http://www.research.ibm.com/software/IBMResearch/multimedia/IJCNN2013.neuron-model.pdf
> http://www.research.ibm.com/software/IBMResearch/multimedia/IJCNN2013.corelet-language.pdf
> http://www.research.ibm.com/software/IBMResearch/multimedia/IJCNN2013.algorithms-applications.pdf
> 
> 
> On Thu, Aug 8, 2013 at 12:15 AM, Steven Oberlin <[email protected]> wrote:
>> You guys see these?
>> 
>> 1) Video:  Sleep as a way to restore/maintain neural plasticity; a lesson 
>> here about the potential value of adjusting pegged permanence values back 
>> toward threshold from time to time?
>> 
>> http://www.scientificamerican.com/article.cfm?id=sleep-brains-way-staying-balance-video-giulio-tononi
>> 
>> 2) Video:  Researchers publish papers in Nature on "connectome" of portions 
>> of rat and fruit fly retinas and another describing decoding motion 
>> detection specialization of fruit fly neurons.  Jeff has blogged about 
>> sparse lessons to be learned from merely mapping connections.  Makes for 
>> pretty pictures, but perhaps better illustrates daunting complexity arising 
>> from a paucity of theory.  The motion detection paper seems to offer more, 
>> but I don't have a Nature subscription.
>> 
>> http://www.nature.com/news/visual-neurons-mapped-in-action-1.13520
>> 
>> 3) Video:  IBM researchers brainstormed applications of low-power, 
>> high-density neuromorphic silicon, funded by DARPA SyNAPSE program.  (I 
>> found myself trying to read the displays in the background that the 
>> presenter *wasn't* describing…)
>> 
>> http://www.engadget.com/2013/08/08/ibm-research-reveals-new-silicon-chip-foundation
>> 
>> -Steve O.
>> _______________________________________________
>> nupic mailing list
>> [email protected]
>> http://lists.numenta.org/mailman/listinfo/nupic_lists.numenta.org
> 
> _______________________________________________
> nupic mailing list
> [email protected]
> http://lists.numenta.org/mailman/listinfo/nupic_lists.numenta.org
_______________________________________________
nupic mailing list
[email protected]
http://lists.numenta.org/mailman/listinfo/nupic_lists.numenta.org

Reply via email to