Biomedical Engineering Reference
In-Depth Information
Within this framework, the time taken to reduce C below a threshold was
computed as a function of B , the number of bursts in HVC per motif. The
result was that the learning speed decreased dramatically as a function of B ;
this was found heuristically to be a consequence of the increasing interference
of weight updates for different synapses.
8.8 Lights and Shadows of Modeling Brain Activity
The brain is probably the most complex “device” found in nature. Are we
ever going to be able to model its dynamics? The brains of songbirds are a
tempting field. The number of nuclei involved in the production of song is
relatively small, as well as the number of nuclei constituting the “learning”
pathway. Our knowledge of the roles played by these nuclei is growing rapidly,
as well as our knowledge of some detailed physiological features of the neurons
within them.
On the other hand, thousands of nonlinear units coupled through a breath-
taking number of connections are the physical substrate of this “device”. How
complex should our description of an individual unit be? How many units are
a reasonable number for an emulation of the activity of a nucleus? Our discus-
sion of the complexity displayed by the physical apparatus, when nonlinear
effects are taken into account in models involving a few variables, gives us
an intuition about the di culties of this program. There are no easy an-
swers to these questions. There is a wide consensus that conductance models
(such as those proposed by Hodgkin and Huxley) are an appropriate point to
start. Dynamicists fight for the hypothesis that simpler models that capture
basic dynamical properties of these equations [Izhikevich 2005] can be used
instead of complex conductance-based models, and some physicists are test-
ing this hypothesis by replacing real neurons by hardware devices obeying
simple dynamical rules [Szucs et al. 2000, Aliaga et al. 2003]. The problem of
the proper dimensionality in which to study these issues is not much easier.
There is no qualitative theory of extended nonlinear problems. On the other
hand, in many cases, we can try to build a consistent story with data from a
few neurons.
Quantification in this discipline will not be an easy adventure. Other dis-
ciplines, such as physics, have built confidence in general laws that allow us to
generate knowledge by theoretically exploring their consequences. Moreover,
it is always possible in physics to isolate a phenomenon and study it under
the simplest possible conditions. None of these basic pillars is available to the
modeler in theoretical neuroscience. However, it is the overwhelming interest
of the questions that keeps research in this field alive.
Search WWH ::




Custom Search