Symposium

Mini-Symposium DEC/IBENS

Practical information
15 April 2019
10:30am-12:30
Place

ENS, Jaurès, 24 rue Lhomond, 75005 Paris

LSP

Mini-Symposium organisé par Pascal Mamassian (LSP/DEC) et Jean-François Léger (IBENS).

10:30am - 11:30am Frédéric Chavane (Institut des Neurosciences de la Timone, Marseille)
The role of cortical waves in shaping the dynamic processing of visual information

Since the pioneering work of the Hubel and Wiesel, the visual system is mostly conceived as a feed-forward hierarchical flow of sensory information. Accordingly, low-level visual information (such as position and orientation) is extracted locally within stationary receptive fields and is rapidly cascaded to downstream areas to encode more complex features. Such a framework implies that processing at each level of the visual system must be fast, efficient and mostly confined to network of neurons with overlapping receptive fields. In recent work, however, we have demonstrated that any local stationary stimulus is, in itself, generating waves propagating within each cortical steps of visual processing. Visual information thus does not stay confined to a particular retinotopic location but instead invades a large cortical territory, connecting neurons with neighboring receptive fields. What could be the computational advantage of cortical waves in the processing visual information? I will show that, in response to a non-stationary sequence of visual stimuli, such as an object moving along a trajectory, these waves interact non-linearly with feedforward and feedback streams. They hereby shape the representation of moving stimuli within cortical retinotopic maps to encode accurately the object velocity. I will propose that propagation of intra-cortical wave can subtend generic computations by which the visual system keep track of moving object and generates accurate and unambiguous predictions.  

11:30am - 12:30 Matteo Carandini (University College London, UK)
Brain-wide activity in a decision based on vision and value

Behavior arises from neuronal activity patterns that are potentially distributed across multiple brain regions, representing computations that may be hidden, i.e. not directly related to motor outputs. Revealing its basis, therefore, requires neuron-level recordings across the brain, causal manipulations, and models that capture the resulting data. We are following this approach to understand the behavior of mice that report decisions based on vision and value. While mice performed this task, we inactivated individual areas of cortex, we modulated the activity of dopaminergic neurons, and we recorded from over 30,000 individual neurons. By using simple models we were able to describe the neural activity in terms of hidden computations, and predict the effects of inactivations. The results point to distributed representations of vision, decision, and action, and to simple neural computations that estimate value and mediate learning.