This is a fundamentally different approach to neuromorphic computing than Numenta's, but it's still neuromorphic computing and that's awesome!
The reason this report is interesting is the power consumption they report. Their chip consumes around 63 mW (that's milli-Watts) compared with a thousand times that for a conventional microprocessor. They claim it can do real-time multi-object recognition in a 400 by 240 pixel video stream which you can't do with a single conventional microprocessor so on tasks like this their power consumption is more than 1000-fold better than conventional microprocessors. Here is my fair-use quote from the paper: "At the operating point where neurons fire on average at 20 Hz and have 128 active synapses, the total measured power was 72 mW (at 0.775 V operating voltage), corre- sponding to 26 pJ per synaptic event (consid- ering total energy). Compared with an optimized simulator (25) running the exact same network on a modern general-purpose microprocessor, TrueNorth consumes 176,000 times less energy per event (supplementary section S12). Compared with a state-of-the-art multiprocessor neuromor- phic approach (13) (48 chips each with 18 micro- processors) running a similar network, TrueNorth consumes 769 times less energy per event (supple- mentary section S12)." On Sun, Aug 10, 2014 at 8:25 AM, cogmission1 . <[email protected]> wrote: > It seems to me the only reason to get excited about this is if the chip can > host CLA algorithmic units...? Is this everyone else's take, or am I missing > something? I guess the fact that people see this as a valid expenditure of > effort brings legitimacy to HTM's but aside from that? > > David > > > On Sun, Aug 10, 2014 at 10:07 AM, cogmission1 . <[email protected]> > wrote: >> >> ^__^ >> >> >> >> >> On Sun, Aug 10, 2014 at 9:56 AM, Jim Bridgewater <[email protected]> >> wrote: >>> >>> As someone who has read a lot of academic papers I can say with full >>> confidence that none of them are worth $20. >>> >>> On Sun, Aug 10, 2014 at 2:58 AM, cogmission1 . >>> <[email protected]> wrote: >>> > Is it worth the $20 to read the article? Is the chip something that >>> > could >>> > host a CLA? >>> > >>> > David >>> > >>> > >>> > On Sat, Aug 9, 2014 at 11:58 PM, Matthew Taylor <[email protected]> >>> > wrote: >>> >> >>> >> This project is not associated with Winfried's team or HTM. >>> >> >>> >> Sent from my MegaPhone >>> >> >>> >> On Aug 9, 2014, at 7:44 PM, "cogmission1 ." >>> >> <[email protected]> >>> >> wrote: >>> >> >>> >> Why is there no mention of Numenta? Is the algorithmic processing on >>> >> board >>> >> the chip not related to HTMs at all? >>> >> >>> >> >>> >> >>> >> >>> >> On Sat, Aug 9, 2014 at 9:09 PM, Jim Bridgewater <[email protected]> >>> >> wrote: >>> >>> >>> >>> I'm guessing it's a different team since Winfried is not on their >>> >>> Science paper http://www.sciencemag.org/content/345/6197/668.abstract >>> >>> nor is he in the photograph in the article you linked to. In >>> >>> addition >>> >>> to that their computer uses spikes for inputs and outputs and >>> >>> Winfried >>> >>> has been a skeptic of that approach in the past, I believe it was in >>> >>> one of the videos from the conference at Sandia this spring that I >>> >>> heard him voice this opinion. If that's true then IBM has at least >>> >>> two teams working on this which is great. >>> >>> >>> >>> >>> >>> >>> >>> On Sat, Aug 9, 2014 at 6:46 PM, Erik Blas <[email protected]> >>> >>> wrote: >>> >>> > I wonder if Winfried is working with this team? >>> >>> > http://www.research.ibm.com/articles/brain-chip.shtml >>> >>> > >>> >>> > Programming language for the architecture: >>> >>> > >>> >>> > >>> >>> > https://dl.dropboxusercontent.com/u/91714474/Papers/020.IJCNN2013.Corelet.pdf >>> >>> > >>> >>> > I'm excited to see the idea of using a collection of prediction >>> >>> > cores >>> >>> > as >>> >>> > primitives to build with, an idea I've bantered around with some >>> >>> > (though >>> >>> > I've been fixated on the primitive being an abstraction for brookes >>> >>> > modules >>> >>> > informed by nupic predictions and the inputs passed into the models >>> >>> > for >>> >>> > said >>> >>> > predictions). Exciting times are ahead! >>> >>> > >>> >>> > _______________________________________________ >>> >>> > nupic mailing list >>> >>> > [email protected] >>> >>> > http://lists.numenta.org/mailman/listinfo/nupic_lists.numenta.org >>> >>> > >>> >>> >>> >>> >>> >>> >>> >>> -- >>> >>> James Bridgewater, PhD >>> >>> Arizona State University >>> >>> 480-227-9592 >>> >>> >>> >>> _______________________________________________ >>> >>> 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 >>> >> >>> > >>> > >>> > _______________________________________________ >>> > nupic mailing list >>> > [email protected] >>> > http://lists.numenta.org/mailman/listinfo/nupic_lists.numenta.org >>> > >>> >>> >>> >>> -- >>> James Bridgewater, PhD >>> Arizona State University >>> 480-227-9592 >>> >>> _______________________________________________ >>> 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 > -- James Bridgewater, PhD Arizona State University 480-227-9592 _______________________________________________ nupic mailing list [email protected] http://lists.numenta.org/mailman/listinfo/nupic_lists.numenta.org
