"Populations of neurons in the cerebral cortex represent sensory information, motor commands and other cognitive functions in their seemingly stochastic activity. Population coding works by distributing the information over the spiking activity of many neurons because single neuronal activity is very unreliable. The statistical structure of the variability across the neural population, i.e. the correlations among neurons, has an enormous impact on the encoding of information in the activity of a neuronal population.Because correlations are viewed as a consequence of shared inputs between nearby neurons, the stochasticity is thought to be an inevitable consequence of the hard-wired connectivity and to limit the efficiency of sensory coding. We have recently shown that recurrent neural networks can generate an asynchronous state with arbitrarily low correlations despite large amounts of shared input (Renart, de la Rocha et al 2010). This implies that correlations are not necessarily the consequence of shared inputs and that encoding in the brain need not be intrinsically stochastic.Our goal is to investigate the origin and functional consequences of the observed cortical stochasticity by studying: (1) the relation between brain state, circuit dynamics and correlations and (2) by quantifying their impact on sensory information representation. We will combine recordings of the spiking activity of large cell populations (50-150) in vivo, with the analysis of computational network models. We will characterize the statistics of spontaneous and stimulus-evoked activity in auditory cortex of (1) anesthetized rats and (2) freely moving rats performing a sensory discrimination task. We will quantify the encoding efficiency of auditory circuits across brain states and will develop a computational model to provide a mechanistic understanding of the data. The results will help to elucidate the neuronal correlates of auditory perception and the basis of a neural code."