Rajiv Narayan <rn@bu.edu>

Seminar, Friday, March 5, 2004
<p>
Discrimination of Natural Sounds in the Songbird Auditory Forebrain
<p>
Abstract:
<p>
Although complex natural sounds such as human speech and birdsongs
display dynamic patterns on multiple timescales, conventional methods
for characterizing neural discrimination typically neglect spike train
dynamics. Here I describe a computational model to investigate how the
dynamic structure of spike trains contributes to the discrimination of
natural sounds, and characterize changes in discriminability as a
function of time.
<p>
The model is based on the spectral temporal receptive field (STRF), a
quantitative description of the stimulus-response properties of
auditory neurons. Experimentally derived STRFs from field L (the avian
analog of auditory cortex) are used to model neural spike trains in
response to an ensemble of birdsongs. The discriminability of spike
trains elicited by different songs is then quantified using the spike
distance metric, a recently proposed theoretical measure that is
sensitive to the dynamic structure in spike trains (M.C.W. van Rossum,
Neural Computation vol. 13, 751-763, 2001).
<p>
The results indicate a significant improvement in the discriminability
of songs when the temporal variations of spike trains are taken into
account. I also examine how discriminability evolves over time by
quantifying the effect of increasing the duration of the spike
train. Finally, I assess how changes in parameters of the model STRF
affect the reliability and time-course of discrimination.