Population codes, correlations and coding uncertainty
Andreas S. Tolias 1 and Alexander Ecker 2 and Georgios A. Keliris 2 and Fanis Panagiotaropolulos 2 and Stefano Panzeri 3 and Nikos K. Logothetis 2
1Department of Neuroscience, Baylor College of Medicine, Biological Cybernetics Max-Planck-Institute, 2Biological Cybernetics Max-Planck-Institute, 3University of Manchester,

Despite progress in systems neuroscience the neural code still remains elusive. For instance, the responses of single neurons are both highly variable and ambiguous (similar responses can be elicited by different types of stimuli). This variability/ambiguity has to be resolved by considering the joint pattern of firing of multiple single units responding simultaneously to a stimulus. Therefore, in order to understand the underlying principles of the neural code it is imperative to characterize the correlations between neurons and the impact that these correlations have on the amount of information encoded by populations of neurons. We use chronically implanted tetrode arrays to record simultaneously from many neurons in the primary visual cortex (V1) of awake, behaving macaques. We find that the correlations in the trial-to-trial fluctuations of their firing rates between neurons under the same stimulation conditions (noise correlations) in V1 were very small (around 0.01 in 500 ms bin window) during passive viewing of sinusoidal grating stimuli. We are also measuring correlations in extrastriate visual areas and investigating the impact of correlations on encoding stimulus uncertainty by neuronal populations, under different stimulus and behavioral conditions.