Marcos, So are you saying this is yet another example of the "new phrenology" like the MRI maps that show this and that lighting up when you do such and such? It can also be asked, how useful is that? I'm actually not that critical, since I really don't know.
I just went to a workshop where there were jaw-dropping amazing visuals and mappings of mouse neurons. There are just tons of complexity in the connectome. The workshop was about how to get findings from neuro-biology into the hands of machine learning people. The neuro folks talked in the morning and the machine folks in the afternoon and although there was a desire to bridge I think the feeling was that this was going to be very hard to do. But that is not to say that it isn't worth the effort. Incidently, (and I hope that I didn't pluck this from this list earlier) but recently this came out that illustrates how hard it is to infer function from internal measurements: Neural System prediction and identification challenge http://www.ncbi.nlm.nih.gov/pmc/articles/PMC3872335/ Tom >Message: 5 >Date: Wed, 12 Feb 2014 09:39:04 -0500 >From: Marcos Canel <[email protected]> >To: "NuPIC general mailing list." <[email protected]> >Subject: Re: [nupic-discuss] Intelligent Chemicals >Message-ID: > <CAKa7D7prJAGhf08behaY41o9H1=+y7p63gs4+p4sdzdjdd3...@mail.gmail.com > >Content-Type: text/plain; charset="iso-8859-1" > >Another reminder that algorithms can be rendered in a variety of mediums. > >Interesting though, that one might know everything about the medium in >question yet still know relatively little about the algorithms it >implements. You could imagine a chemist studying for years all of the >different species involved in this system and still drawing a blank as to >what it is exactly that the system does. I wonder if neuroscience isn't in >a similar rut when it comes to the brain. > >I think a lot of people have the idea that the holy grail of neuroscience >would be to have a single-neuron resolution brain imaging technology that >could be used to track signal processing within the brain from the moment >light hits the retina all the way to behavioral output/long-term memory >storage/what have you. But where does such a technology actually put us as >far as understanding how the brain works? > >I read an interesting paper about imaging the neurocircuitry of classical >conditioning in Zebrafish[1], but was surprised at how seemingly little >actionable information there was from an engineering point of view, despite >the group's managing to precisely track learning to a relatively small >cluster of neurons. The techniques used in the study are very interesting >and it's definitely worth the read, but I wonder how long it will be till >such research produces ideas for the data structures and algorithms >actually used within the brain, rather than notions of approximately where >and how learning takes place. > >My concerns are probably overblown due to lack of familiarity with the >literature, but it seems like some groups in neuroscience forget Feynman's >adage that what one cannot build, one does not understand. > >[1] > http://download.cell.com/neuron/pdf/PIIS0896627313003115.pdf?intermediate=true > > >On Fri, Feb 7, 2014 at 7:22 PM, Fergal Byrne <[email protected] >wrote: > >> +1 Jeff. It's taken half a billion years to evolve the machinery to turn >> chemical reactions into intelligence, and we're only starting to figure it >> out in the last few decades. >> >> >> On Sat, Feb 8, 2014 at 12:00 AM, Jeff Hawkins <[email protected] >wrote: >> >>> This is pretty cool. However, I don't think this will be a practical way >>> to >>> implement HTMs. The mechanisms for HTM require many steps of processing >>> and >>> much of it is locally specific, e.g. each dendrite segment. I don't see >>> how >>> this could be done with chemical gradients. Would love to be proved >>> wrong. >>> Jeff >>> >>> -----Original Message----- >>> From: nupic [mailto:[email protected]] On Behalf Of Chris >>> Jernigan >>> Sent: Thursday, February 06, 2014 4:08 PM >>> To: NuPIC general mailing list. >>> Subject: Re: [nupic-discuss] Intelligent Chemicals >>> >>> I forgot to ask, do you think this could be used to implement an efficient >>> HTM? >>> >>> On Feb 6, 2014, at 6:57 PM, Chris Jernigan >>> <[email protected]> wrote: >>> >>> > They are doing some really cool research at Harvard on using chemical >>> reactions to create intelligent systems. Here is the article: >>> http://www.seas.harvard.edu/news/2013/12/programming-smart-molecules >>> > >>> > And a PDF of the original paper here: >>> >>> http://papers.nips.cc/paper/4901-message-passing-inference-with-chemical-rea >>> ction-networks.pdf< http://papers.nips.cc/paper/4901-message-passing-inference-with-chemical-reaction-networks.pdf > >>> > >>> > Most of the math is way beyond me, but the concept seems solid enough. >>> What do you guys think? >>> > >>> > -Chris
_______________________________________________ nupic mailing list [email protected] http://lists.numenta.org/mailman/listinfo/nupic_lists.numenta.org
