 |
 |
|
|
B.D. Wright, K. Sen, W. Bialek and A.J. Doupe
Discovering features of natural stimuli relevant to songbird auditory neurons
University of California, San Francisco
One way to determine the features of sensory stimuli important to a neuron is to compute a spike-triggered average of some representation of the stimulus. In the case of the auditory system, one often constructs the spectro-temporal receptive field. However, this method produces only one average stimulus feature to which the neuron is sensitive, and describes at best the linear component of the stimulus-response relationship, while auditory neuron responses can be highly nonlinear. Here we apply and develop a method in which we use the neuronal response to extract, from a large stimulus space, multiple auditory features to which the cell is sensitive.
The method expands upon techniques first used in characterizing motion sensitive neurons in the blowfly. We compute the covariance matrix, C, of the stimulus in a window surrounding a spike and, to account for intrinsic correlations in our natural stimulus ensembles, a corresponding prior covariance matrix, C_prior, of the stimulus. If the neuron is sensitive to only a small number of stimulus dimensions or features, then C - C_prior will have mostly zero eigenvalues; eigenvectors associated with nonzero eigenvalues provide a low dimensional space of relevant stimulus features. We use this method to characterize the neuronal responses of cells in the auditory field L complex of the zebra finch, using stimuli consisting of recordings of birdsong. Preliminary results indicate that from a stimulus space with roughly 1000 dimensions, only a few are relevant to the response of neurons in field L. We explore these dimensions, using probabilistic and information theoretic techniques, to determine potentially nonlinear interactions among them.
|
|
 |
|
Saturday, December 21, 2024
|
| |
The Swartz Foundation is on Twitter: SwartzCompNeuro
|
|
|
 |
print