To kickstart 2017, let me invite you to a seminar lecture by Gergo Orban (Wigner Research Center for Physics, Hung. Acad. Sci.) to be held in the KOKI lecture room on 5 Jan at 11 am. Orban’s group aims to understand the computational principles underlying sensory processing. The speaker is a Lendulet Research Fellow and recently published an insightful paper on the role of variability in visual processing that incited many news pieces (see e.g. http://mta.hu/tudomany_hirei/valosag-es-illuziok-kiderult-hogy-miert-a-zajos-agy-a-hatekony-agy-106981, in Hungarian; original paper: Orban, Berkes, Fiser, Lengyel, 2016, Neuron). The title and abstract of his talk:
Sampling-based computations link variability, correlations, and spontaneous activity in the visual cortex
Stimulus unrelated activities — variability, correlations, and spontaneous activity — are prevalent in the nervous system and pose a critical challenge to an in-depth understanding of the processes occurring in the brain. We will discuss how these different forms of activities can be interpreted in the same mathematical framework: probabilistic computations with probabilistic sampling. In particular, we investigate these forms of activities in the visual system and demonstrate that a wide range of experimental data available in the literature on neural variability and covariability recorded from various model systems (mice, cats, ferrets, and monkeys) have a unified interpretation under stochastic sampling. Further, considering computations in a hierarchical system we formulate predictions on the stimulus-dependence of noise correlations in the primary visual cortex. In order to test these predictions, we set up an experiment with awake behaving monkeys. Novel multiunit recordings from three experiments are presented that corroborate our predictions and provide evidence that stimulus structure is informative with respect to the hierarchical aspects of neural computations. We argue that variability and correlations provide insight into how neural circuits implement optimal computations.