[Probcogsci] Computational neuroscience talk: Wed (03/04) 2 PM, Natural Computation Lab Seminar Room (SSRB floor 2)

[Nabil.Bouaouli at ens.fr]

A Probabilistic Framework to go from Function to Circuitry: A View  
from Change Detection
Nabil Bouaouli
Group for Neural Theory
ENS-Collège de France

The brain continuously processes the signals that emerge from the  
dynamic
sensory environment. A well admitted view about the strategy applied  
by the
nervous system is that it analyses the “sensory scene” as an ensemble  
of basis
features which constitute the signal (e.g. colors, orientations,  
tones).  Each
feature, then, activates specific sensory neurons which are sensitive  
to its
appearance. On the other hand, theses neurons are known to respond more
strongly to transient than to steady stimuli.  In this work, we  
hypothesize
that they are suited to efficiently detect changes in the environment  
(i.e. the
appearance of their preferred features) as soon as these changes occur  
and the
question we want to answer is how are sensory stimuli detected and  
coded and
what are the mechanisms involved  in this process.
To address this issue, we provide a probabilistic framework of change  
detection
that goes from functional level to cellular level. Namely, requiring the
detection of fast and unpredictable changes in the environment, the  
model
predicts the likely sensory circuits needed to achieve this goal. This  
(micro)
circuit involves parallel feedforward excitatory and “delayed”  
inhibitory
pathways and predicts the short-term temporal dynamics of the synapses
involved. Moreover, it shows how the biophysical properties of these  
synapses
could be shaped by the statistics of the environment. In terms of  
sensory
coding, our model exhibits naturally adaptive behavior as a built-in  
process.
In response to changes in the level of stimulus fluctuations, the  
model adapts
its sensitivity in qualitatively the same way as retinal and LGN cells  
do.  As
a key feature of this model is a “bounded” and delayed inhibition that  
depends
on the temporal statistics of the environment, we propose that  
feedforward
inhibition may play a general role in sensory processing over many  
timescales
and argue that the balance between excitation and inhibition may have  
direct
implications on the efficiency of early sensory processing.