Laboratoire de Neurosciences cognitive, ENS-INSERM, France
Wednesday 13 January 2010
Seminar Room B10 (Basement)
Alexandra House, 17 Queen Square, London, WC1N 3AR
Bayesian inference with spikes. Implication for the neural code, sensory processing and working memory.
The Bayesian framework provides normative models that proved powerful to describe human behavior. However, the neural basis of probabilistic computations remain largely unknown. Like behavioral tasks, many elementary problems solved by neural structure can be formalized as finding the underlying causes for the sensory observations.
We will show that the integrate and fire dynamics of biological neurons and the statistics of their spike trains suggest that single spikes represent sudden increase in the (log) probability of a variable. We will next explore the implication of this hypothesis for the dynamics of single neurons, single synapses, and population of neurons.
Some of the derived models are familiar, in the sense that they converge to pre-exiting computational or biophysical models. Others bring new concepts of neural mechanism and alternative views of the neural code. Overall, this approach accounts for the highly non-invariant and adaptive nature of sensory responses, tuning curves and receptive fields. It also resolves an apparent paradox: why the brain can compute so reliably (sometimes) with spiking neural responses that are so variable.