Response Variability in LGN

M.Eisele
coauthors: C.Weng, J. Z. Jin, C. I. Yeh, J.-M. Alonso, K.D. Miller   

Ken&Michael: Center for Theoretical Neuroscience,  Center for Neurobiology and Behavior, Columbia University
 
Other authors: Department of Biological Sciences, College of Optometry, State University of New York
New York, NY, USA

Neural response properties vary not only between cell classes, but also, more gradually, within each class. Some of these variations presumably help in processing the diversity of sensory stimuli, while others are just biological noise. We argue that response variations need to reach a certain size before they become useful for signal processing and that this can be used to distinguish them from smaller variations that are just noise. We demonstrate how to make this distinction in the lateral geniculate nucleus (LGN) of the cat. Receptive fields of X- and Y-cells were mapped in space and time using reverse correlation with dense white noise stimuli of high or low contrast or with sparse noise. These receptive fields were then analyzed with a modified version of principal component analysis (PCA). Apart from the obvious diversity of preferred spatial positions, the diversity response time-courses also seems large enough to play a functional role, while other forms of diversity are much weaker and probably just noise. This result agrees well with what we know about how LGN outputs are processed in primary visual cortex. By comparing the diversity in LGN to the responses of simple cells, which were recorded under similar conditions, we find that the actual diversity in LGN is near the optimal diversity for cortical signal-to-noise. These results show how gradual, apparently random variations can play a role in optimal encoding.