About the presenter:
Basic Neuroscience Department University of Geneva, Andrew Carnegie Prize in Mind and Brain Sciences
Most computations performed by the brain are subject to large amount of uncertainty because sensory inputs are noisy and ambiguous. As Laplace and Hemlholtz have pointed out over the last two centuries, the best way to compute in the presence of uncertainty is to adopt a probabilistic approach, that is, to represent knowledge in the form of probability distributions and to perform probabilistic computations. Several hypotheses have emerged recently regarding the neural implementations of these probabilistic inferences. I will review in particular one such hypothesis, based on the notion of probabilistic population codes, which shows how neurons perform probabilistic inference with simple biologically plausible linear and nonlinear neural circuits. We are applying this approach to a wide array of seemingly different behaviors such as decision making, visual search, simple arithmetic, perceptual learning, multisensory integration and olfactory processing to name a few. Interestingly, the mechanisms involved are so simple that they might also be used in insects, suggesting that probabilistic inference provides a general framework to understand neural computation in all species.