[Comp-neuro] Hilbert's questions

Wei Ji Ma wjma at cpu.bcm.edu
Wed Aug 13 20:58:58 CEST 2008

I agree with Randy on the importance of also having a top-down strategy,
and feel that this view is unfortunately underrepresented among
computational neuroscientists. It's wonderful to build large networks of
biologically realistic neurons, but if you get too absorbed in the
details on that end, it's easy to forget that there is a wealth of
psychophysics data waiting to be explained in neurobiological terms.
Stronger yet, I would say that at least for most problems in perception,
your network is useless unless you do exactly that explaining. 

I don't mean plain parameter fitting - if your network has enough bells
and whistles, you can make it do anything. What I think constitutes real
understanding in computational neuroscience is a principled theory of
neural coding that can link neural activity (rates or times, whichever
you prefer) to behavioral quantities. Eventually it's nice (and maybe
even necessary) to have a network of spiking neurons that can
corroborate your high-level conclusions, but the centerpiece should be a
mathematically rigorous connection between neuronal populations and
behavior. (And it's looking more and more like this should be formalized
in terms of probability distributions over stimuli.) 

Frankly, I don't even care that much where exactly in the brain you are
modeling the neurons. Whatever the behavior, it has to be happening
somewhere, and if that is in area X or Y or somehow distributed across
them is less interesting than what computation the neurons are

Wei Ji (Whee Ky) Ma, Ph.D.
Assistant Professor
Department of Neuroscience
Baylor College of Medicine
Houston, TX 77030
Phone: 1-713-798-8407

-----Original Message-----
From: comp-neuro-bounces at neuroinf.org
[mailto:comp-neuro-bounces at neuroinf.org] On Behalf Of Randall O'Reilly
Sent: Tuesday, August 12, 2008 12:19 PM
To: james bower
Cc: comp-neuro at neuroinf.org
Subject: Re: [Comp-neuro] Hilbert's questions

There has been a remarkable absence of consideration for the  
functional level of the brain: i.e., the huge field of cognitive  
neuroscience, in this discussion and in the field of computational  
neuroscience in general.  There is no way you are going to figure out  
something as complex as the brain using a purely bottom-up strategy.   
If you look at the work in computational *cognitive* neuroscience,  
where the behavioral level of analysis is taken seriously, the picture  
is much less bleak than the assessments provided here.  There are many  
models that relate the biological properties of neurons and brain  
areas to cognitive functions associated with those brain areas, and do  
a very good job of capturing a large proportion of the variance of  
both levels.

For example, the hippocampus is essentially a "solved problem" in  
terms of the general framework for how its biological properties  
enable its well established role in memory.  Recent work by Tonagawa's  
group and several others have verified the predictions from a number  
of generally convergent computational models, regarding the specific  
roles of areas CA3, DG, CA1, etc.  By capturing the most global, high- 
level properties of the system first, across both behavior and  
biology, these models provide a framework within which more detailed  
questions and models can be developed.

In almost every domain, a hierarchical coarse-to-fine strategy is the  
most efficient way to understand something.  First you figure out the  
most basic properties of the system, and then you fill in the  
details.  Some would argue that this is not possible in the brain, but  
I think the existing work already refutes that argument.  People who  
remain fixated on individual neurons and synapses may not appreciate  
this, and regard the system as a huge unsolved puzzle, but this is  
just because they are so zoomed in on the details that they are  
missing the big picture.  The big picture is filling in quite  
rapidly.  There are similarly successful models for prefrontal cortex,  
basal ganglia, and sensory neocortex, etc.

Taken together these models strongly suggest that, to understand how  
the system actually functions, you don't need to simulate every last  
detail of a neuron, nor its connectivity.  Certain details are rather  
important (e.g., inhibition in the hippocampus is critical for  
establishing a sparse distributed representation, which minimizes  
interference and enables episodic memory, place cells, etc), but in  
general a fairly simple "integrate and fire" model of the neuron is  
sufficient to capture a large portion of the functional variance of  
what the brain actually does.

Of course, I can't "prove" any of these assertions to the satisfaction  
of all skeptics, and I'm rather an optimist overall, but I think this  
field is definitely missing out on the big picture.  Certainly there  
is a huge amount still unknown, but if you squint your eyes just  
right, I think the picture is filling in quite nicely..

- Randy

On Aug 12, 2008, at 9:47 AM, james bower wrote:

> I would say for sure that individual neurons are communicating --  
> just that communication is not dependent on any individual neuron  
> (in mammals), nor can one understand what they are communicating  
> independent of the population - a nice enigma.
> With respect to wiring - 'we' believe that nervous systems represent  
> what they know in their wiring -- 'we' also believe that the  
> modification of wiring takes place at the level of individual  
> neurons (and even synapses).  So for didactic purposes:
> 1) does the function of an individual brain depend on the detailed  
> wiring of that brain (likely)
> 2) can we therefore understand how brains function in general, by  
> working on multiple individuals let alone multiple species
> 3) in other words, what level of wiring specification do we need?
> and do we have the patience?
> Speaking of grubby, now probably mostly lost in history, the first  
> ''realistic network" modeling effort I ever saw presented was of the  
> sea slug tritonia, by an engineer (MIT) turned serious  
> experimentalist, Peter Getting - Neuroscience 1981, I think.
> Peter Getting had originally taken a faculty position at Stanford,  
> and, on the assumption that wiring was everything, set about trying  
> to understand the connections between the few (I think 6) types of  
> motoneurons that control the swimming (if you want to call it that)  
> of Tritonia.  Problem was that after his 6 year junior faculty  
> appointment, he had only completed characterizing 3 of the 6 (as I  
> remember) sets of connections.  This was not deemed reasonable  
> progress, he was denied tenure and ended up taking a position in  
> Iowa, where he steadfastly continued to complete the circuitry.  He  
> did, and presented the results at the neuroscience meeting -- and I  
> remember being astounded.  Peter would have actually been a major  
> part of the first course in woods hole, had he not had a massive  
> stroke while running (which he did many miles per day), ending up  
> incapacitated.
> However, for sure, we now know from invertebrate systems that the  
> individual connections of individual neurons within an individual  
> matter --
> So -- if one believes in the importance of wiring -- shouldn't we  
> all be working in invertebrate preparations?
> Not an entirely rhetorical question -- it is clear from the history  
> of science in general and physics in particular that picking the  
> right problem is a key to progress.  Thus, Newton 'discovered' the  
> inverse square law by examining the (nearly) circular orbit of the  
> moon around the earth, for which he also had much better distance  
> data, rather than looking at the sexier elliptical movements of the  
> planets around the sun.
> Maybe we should give up on cerebral cortex for several hundred years  
> and all study tritonia instead.
> Jim
> On Aug 12, 2008, at 10:08 AM, Bill Lytton wrote:
>>> grandest level it seems to me there is only one question: "What is  
>>> each neuron communicating,
>>> and how is the message encoded."
>> I thought it was noted in recent discourse that the answer is  
>> nothing and not? -- ie
>> populations are needed.
>>> That said, it could be an interesting exercise to come up with a  
>>> list of the current Top Ten
>>> Topics attracting the attention of
>> Personally I would echo Martin and Douglas in their many papers  
>> (from which I recommend Neuron
>> 2007 56:226-238 for its broad scope) that we need to know how it is  
>> wired where 'it' may be
>> neocortex, thalamus, olfactory cortex or even bug whateveritis-ex.
>> Framed computationally this could involve wiring exploration (which  
>> we are doing lately) or
>> development algorithms or new Hebb variants.  Of course, without  
>> the accompanying
>> physiological/anatomical exploration this will be meaningless.
>> Admittedly this is a rather low-level question without the grand  
>> sweep of a Hilbert q. but
>> then biology is often grubby (even literally) rather than ethereal.
>> Bill
>> -- 
>> William W. Lytton, MD
>> Professor of Physiology, Pharmacology, Biomedical Engineering,  
>> Neurology
>> State University of NY, Downstate Medical Center, Brooklyn, NY
>> billl at neurosim.downstate.edu http://it.neurosim.downstate.edu/~billl
>> ________________________________________________________________
> ==================================
> Dr. James M. Bower Ph.D.
> Professor of Computational Neuroscience
> Research Imaging Center
> University of Texas Health Science Center -
> -  San Antonio
> 8403 Floyd Curl Drive
> San Antonio Texas  78284-6240
> Main Number:  210- 567-8100
> Fax: 210 567-8152
> Mobile:  210-382-0553
> The contents of this email and any attachments to it may be  
> privileged or
> contain privileged and confidential information. This information is  
> only
> for the viewing or use of the intended recipient. If you have  
> received this
> e-mail in error or are not the intended recipient, you are hereby  
> notified
> that any disclosure, copying, distribution or use of, or the taking  
> of any
> action in reliance upon, any of the information contained in this e- 
> mail, or
> any of the attachments to this e-mail, is strictly prohibited and  
> that this
> e-mail and all of the attachments to this e-mail, if any, must be
> immediately returned to the sender or destroyed and, in either case,  
> this
> e-mail and all attachments to this e-mail must be immediately  
> deleted from
> your computer without making any copies hereof and any and all hard  
> copies
> made must be destroyed. If you have received this e-mail in error,  
> please
> notify the sender by e-mail immediately.
> _______________________________________________
> Comp-neuro mailing list
> Comp-neuro at neuroinf.org
> http://www.neuroinf.org/mailman/listinfo/comp-neuro

Comp-neuro mailing list
Comp-neuro at neuroinf.org

More information about the Comp-neuro mailing list