No subject

Dear Connectionnists,


We would like to announce the paper 


"CODING WITH NOISY NEURONS: STABILITY OF TUNING CURVES DEPENDS STRONGLY
ON THE ANALYSIS METHOD" 


by Axel Etzold, Helmut Schwegler, and Christian W. Eurich,


to appear in the "Journal of Neuroscience Methods".


A standard evaluation method for neural data is the construction of a
neural tuning curves. However, the widely used statistical method of
statistical analysis based on on the sample mean and least-squares
approximation for the spike count can perform extremely badly if the
noise distribution is not exactly normal, which is almost never the case
in applications. Here we present a method for constructing neural tuning
curves that is especially suited for cases of high noise and the
presence of outliers. In contrast to traditional methods employing a
point-by-point estimation of a tuning curve, we use all measured data
from all different stimulus conditions at once in the construction.
Using approximation theory, a tuning curve is identified which best
approximates a hypothetical ideal tuning curve across all stimulus
conditions. The influence of several types of noise distributions on the
stability of the parameters of the tuning curve is investigated. A
rank-weighted norm is employed which yields more stable tuning curves
than the traditional least mean squares method and at the same time
conserves information which would be discarded by a median based method.
The theoretical results are applied to responses of cells in rat primary
visual cortex.


A preprint and additional software for MATLAB are available at

http://www.neuro.uni-bremen.de/~web/index.php?id=54&link=/~noisy/noisytuning.html

The Matlab package may be used by everyone to generate tuning curves
automatically from empirical single-cell data.

Best regards,

Axel Etzold