Date: February 07, 2017. 12:00
Location: CCU Seminar Room
Title: From dynamics to computations in recurrent networks.
Affiliation: Ecole Normale Superieure, France.
Recurrent networks of excitatory and inhibitory neurons form the basic computational units in the cortex. Theoretical works have shown that randomly-connected networks naturally generate irregular activity which resembles spontaneous cortical activity. How recurrent networks implement complex, behaviorally relevant computations remains however an outstanding open issue.
In this seminar, I will present two projects addressing this question. In the first part of the talk, I will describe how minimal, low-dimensional perturbations of a random connectivity structure modify the network dynamics and can be exploited to implement specific computations. In the second part, I will examine neural activity recorded in the ferret auditory cortex while animals either passively listened or actively discriminated stimuli. Based on population analyses, I will argue that the type of computation performed in the auditory cortex strongly depends on the behavioral state of the animal.