In cooperation with the Leibniz-Institut für Neurobiologie (Magdeburg) we develop new methods for estimation of auditory spectro-temporal receptive fields (STRF). Instead of simple averaging procedures (spike-triggered average/reverse correlation) we use modern regression and classification methods. The STRFs estimated using these methods produce much better spike rate predictions than the simple methods. 

The linear auditory spectro-temporal receptive field (STRF) is a well-known approach to describe the linear relation between an acoustic stimulus and the response of a neuron evoked by that stimulus. Common STRF estimation methods are reverse correlation – also known as spike-triggered average – and normalized reverse correlation. Currently, we develop and test STRF estimation methods based on statistical learning. Therefore, we show that the linear STRF model
can be reforumlated in terms of linear regression and linear classification [1]. A new method based on support vector machine (SVM) classification is proposed and compared to other linear methods (reverse correlation, normalized reverse correlation, linear regression, Ridge regression, logistic regression). The methods are tested using real spike data from anesthetized gerbils [2] and zebra finches [3] as well as synthetic stimuli (FM-Sweeps, DMR). We used 80% of the data for training and 20% for prediction.

The dimensionality of the data is reduced using principal component analysis (PCA) or regularization techniques. For all stimuli about 5% of the feature dimensions account for about 80% of the total variance.

[1] Arne F Meyer, Max FK Happel, Frank W Ohl and Joern Anemueller: Estimation of spectro-temporal receptive fields based on linear support vector machine classification. BMC Neuroscience 2009, 10(Suppl 1):P147
Poster: