Wednesday 4th January 2017
Ground Floor Seminar Room
25 Howland Street, London, W1T 4JG
Stimulus-dependent modulation of cortical variability: computational function and dynamical mechanisms
Neural responses in the visual cortex are variable, and there is now an abundance of data characterizing how the magnitude and structure of this variability depends on stimulus attributes. These rich data sets allow us to revisit some of our basic assumptions about the functional roles and mechanisms of cortical variability. I will first argue that key aspects of cortical variability and correlations emerge naturally from a top-down theory of cortical dynamics. According to this theory, population activity patterns represent statistical samples from the probability distribution arising from probabilistic inference, and thus their variability directly encodes perceptual uncertainty. Through direct comparisons to previously published data as well as original data analyses, I will show that a sampling-based probabilistic representation accounts for the structure of noise, signal, and spontaneous response variability and correlations in the primary visual cortex. I will then show that a simple model circuit, the stochastic stabilized supralinear network (SSN), featuring strong and fast non-normal amplification and non-linear interactions around a single attractor state, provides a mechanistic account of several key aspects of cortical variability, including the ubiquitously observed variability quenching following stimulus onset and the stimulus tuning-dependence of Fano factors and noise correlations in area V1 and MT. This model represents a qualitatively different regime of cortical dynamics than standard models that either rely on a slow noisy exploration of a multitude of quasi-attractor states, or on chaotic spontaneous dynamics operating in the absence of any attractor states, and the dynamical regime in which it operates may in fact be optimal for sampling-based inference. Taken together, these results allow us to take the interpretation of cortical variability beyond the well-entrenched realm of information transfer and put them centre stage in the study of cortical dynamics and computation.
Joint work with Gergő Orbán, Pietro Berkes, József Fiser, Guillaume Hennequin, Yashar Ahmadian, Dan Rubin, Ken Miller, and Laurence Aitchison.
Máté Lengyel is a Reader in Computational Neuroscience at the Department of Engineering, University of Cambridge, and a research fellow at the Department of Cognitive Science, Central European University. Máté’s interests span a broad range of levels of nervous system organisation, from sub-cellular and cellular through circuit and systems to behaviour. He studies these phenomena from computational, algorithmic/representational and neurobiological viewpoints. Computationally and algorithmically, he uses ideas from Bayesian approaches to statistical inference and reinforcement learning to characterise the goals and mechanisms of learning in terms of normative principles and behavioural results. He also performs dynamical systems analyses of reduced biophysical models to understand the mapping of these mechanisms into cellular and network models. Máté obtained his MSc and PhD at the Eötvös Loránd University, followed by a post-doctoral research fellowship at the Gatsby Computational Neuroscience Unit, UCL, and a visiting research fellowship at the Collegium Budapest Institute for Advanced Study. He has been awarded an Investigator Award by the Wellcome Trust, and more recently a Consolidator Grant by the European Research Council.