[Comp-neuro] "realistic models"

Harry Erwin harry.erwin at sunderland.ac.uk
Thu Aug 21 17:19:59 CEST 2008


On 20 Aug 2008, at 17:09, james bower wrote:

> The issue here is simple - it isn't whether the brain is for sure  
> KCC, it is a question of what assumptions one makes to begin with.   
> In the Marrian case, where one regards the brain as not necessarily  
> the optimal, but one of a number of implementations of a set of  
> algorithms, then one would clearly want to work on the set of  
> algorithms first - and ignore the brain -- the more one suspects  
> that the brain is not just one implementation - but perhaps a KCC  
> implementation -- then one pays more attention to the structure of  
> the brain -- In the end do these two approaches converge -- we can  
> hope -- but I would say they will only converge if the abstract  
> modeler accepts as an assumption (as I know, for example, that Bard,  
> Larry Abbot, and others do), that the brain, in the limit, is the  
> common reference point.

The reason physical science works--Paul Davies suggests--is that the  
aspects of the universe addressed by physical science are not KCC. A  
scientific field that deals with a KCC phenomenon is basically stamp- 
collecting. On the other hand, I believe there are areas of science  
that are materialist but not reducible--what intervenes is Stewart and  
Cohen's 'ant country'. There's reducibility--lawfulness--but it's an  
incomplete theory, and the phenomenon has to be understood at its own  
level, too.

>
> I also completely agree with you on the need to remember that we  
> classify things (neurons, motor-systems, etc), and the brain does  
> not care in the least about our classifications.  Again, in  
> principle, a structurally realistic model of the nervous system also  
> inherently does not have such classifications.  However the more  
> removed your modeling efforts are from that physical structure, the  
> more functionally modular models become.  Something I consider to be  
> a big problem.
>
> I actually think that it is critical to develop a better  
> understanding of what the brain does -- but, my assertion (or going  
> in assumption) that the brain is KCC means that what the brain does  
> is actually reflected fundamentally in its real structure -- another  
> argument for realistic modeling.
>
> I can give you a particular for instance - returning to the  
> cerebellum, looking at its circuitry, Marr / Albus proposed that,  
> basically, the climbing fiber 'taught' the Purkinje cell what  
> pattern of active parallel fibers to recognize.  Thus, the Purkinje  
> cell is seen as a parallel fiber pattern recognizer.  Based on our  
> realistic models and related experimental work, we have provided  
> evidence that the Purkinje cell doesn't even respond to parallel  
> fiber input, and instead, the the parallel fibers, by modulating  
> local regions of the Purkinje cell dendrite, provide contextual  
> information for the Purkinje cells 'evaluation' of inputs it  
> receives from excitatory inputs from synapses associated with the  
> axon of the granule cell as it courses vertically into the molecular  
> layer -- this is a COMPLETELY DIFFERENT type of computation --  
> Although first proposed almost 30 years ago, no current abstract  
> model of cerebellar function even includes this ascending axon  
> input.  When they do, you can bet that it will be included in such a  
> way as to maintain the core notion of Purkinje cell instructive  
> learning.
>
> Jim
>

Let me get this straight--granule cell axons first synapse strongly on  
a single Purkinje cell and then split into parallel fibres, synapsing  
weakly (both directly and indirectly via interneurons) on a large  
number of other Purkinje cells. So the output of a Purkinje cell is  
dominated by a few granule cell axons and is tuned by a large  
population of parallel fibres. How do the climbing fibres fit in?

I understand that a major part of the cerebellum is used to propagate  
motor plans forward (and backwards). How does the cerebellar  
architecture support that function?

--
"If academic research is not devoted to finding the truth, it is a  
form of propaganda, and not necessarily to be preferred to other  
forms, much cheaper and perhaps more persuasive." (Russell 1993)

Harry Erwin





More information about the Comp-neuro mailing list