[Comp-neuro] Hilbert's questions

Axel HUTT axel.hutt at loria.fr
Wed Aug 13 10:38:14 CEST 2008

> [...]
> As far as the grande (sic) questions in neuroscience go, at the  
> grandest level it seems to me there is only one question:  "What is  
> each neuron communicating, and how is the message encoded."
> Once we know that, the rest is clean up.

as a physicist by education but working in neuroscience, I am really 
puzzled about the approach to start from the single neuron to understand
information processing. For me, this seems as if one aims to understand 
water waves, hydrodynamic convection or turbulence in fluids by studying
 the collisions of the fluid molecules or the dynamics of the atoms. 

In physics, many hydrodynamic phenomena are well-described by
 macroscopic equations (e.g. the Navier-Stokes equation). The reason
for the power of the macroscopic equation is the optimal choice of the 
description level. There are also good microscopic models for the
 intra-molecular dynamics of fluids and some for the interaction between
 molecules. However transversing the scales is very difficult, though
 the research starts from each description level (maacro and micro)
to find the bridge between the scales.

In neuroscience, we have a similar problem of scales, while the (living)
 neural systems are complex compared to fluids. However, in
neuroscience the choice of the description level is also very important
 and IMHO one should aim to find reasonable descriptions for each level.
Currently, macroscopic models such as neural population models assuming
 an underlying population rate code allow to reproduce successfully
 rather macroscopic activities (EEG/MEG,LFP). However, much work has to
 be done to include spatial topologies. Further network models
built of spiking neurons (time-coding models) allow for the successfull 
description on the single neuron level. Each spatial/temporal level 
allows for the reasonable description of specific phenomena, e.g. 
macroscopic activity is strongly related to cognitive phenomena and 
some first stages in perception, while microscopic activity may
 reflect the neural performance in time-related tasks and some temporal 
aspects of perception. Hence each description level has its own meaning
and importance and it may be a good idea to keep working on improved the
models for each level and scale before combining the scales. 

In the ongoing discussion, somebody wrote (sorry, do not know anymore
who) that neuroscience research is in some way in the situation where
 physics was in the 16th century. I fully agree. Though we measure
 plenty of data and have new exciting and powerful techniques to measure
 activity in the brain, sometimes we are far from understanding what is 
going on in the brain. Hence I suggest to focus on improving the 
models in the respective scale to describe the respective experimental

Some problems may be:

* how does the spatial topology of visual areas in mammals affect the
processing of visual information ? Are extended rate-coding models
 sufficient to describe the corresponding major underlying mechanisms
including the intra-area connections and the inter-area feedback loops?

* what is an optimal model approach to describe time-coding effects 
in networks including different types of synapses, dendritic trees and
 the axonal branching system ?

Axel Hutt
INRIA CR Nancy - Grand Est
CS20101, 54603 Villers-lès-Nancy Cedex

More information about the Comp-neuro mailing list