Sophie Deneve
http://www.gnt.ens.fr/people.php?id=1
Wednesday 10th April 2013
Time: 4pm
Basement Seminar Room
Alexandra House, 17 Queen Square, London, WC1N 3AR
Biography
Sophie Deneve did a PhD in University of Rochester (2003) and a postdoct in Gatsby Computational Neuroscience unit (2003-2004) before obtaining a faculty position in CNRS, France, in 2004. Since 2006 she co-directs to Group for Neural Theory (Ecole Normale Supérieure, Paris). She obtained a Marie Curie Team of Excellence award (2006), a James McDonnell foundation award (2012) and ERC consolidator grant (2013). Her research focuses on neural coding and in particular how the brains represents and deals with uncertainty and noise.
Learning optimal spike-based representations using predictive coding
Neural networks can integrate sensory information and generate continuously varying outputs, even though individual neurons communicate only with spikes---all-or-none events. Here we show how this can be done efficiently if spikes communicate ``prediction errors'' between neurons. We focus on the implementation of linear dynamical systems and derive a spiking network model and a spike-time dependent learning rule from a single optimization principle. Our model naturally accounts for two puzzling aspects of cortex. First, it provides a rationale for the tight balance and correlations between excitation and inhibition observed in cortical neurons. In fact, learning and maintaining an efficient spike-based code is equivalent to learning and maintaining the balance between excitation and inhibition. Second, this predicts asynchronous and irregular firing as a consequence of predictive population coding, even in the limit of vanishing noise. However, we show that balanced spiking networks have error-correcting properties that make them far more accurate and robust than comparable rate models: the large variability at the level of single neurons cancels at the level of the population. Our approach suggests that spike times do matter when considering how the brain computes, that the neural variability is largely due to degeneracy, not internal noise, and that the reliability of cortical representations have been strongly under-estimated.