A seminar by [Etienne Thoret for the CONECT group.
Etienne Thoret (ILCB/PRISM/LIS/AMU) kindly accepted to present his research project during our novel series of CONECT-core © seminars (=seminars open to all but focused on the core theoretical scientific questions of the CONECT members):
Explainable AI for computational auditory neurosciences
Machine learning and deep neural networks have been raised as compelling models to simulate a broad range of tasks on signals: from classification of sound events to the prediction of human physiological state from electrophysiological data. But what do we really understand about these models and how do they process the information they have been trained to process? As users, we often use them as tools without precisely understanding their mechanistic and representational underpinnings. In this talk, I’ll present recent works on how we can take part of these computational systems to answer fundamental research mysteries on auditory perception, speech production and cerebral processing. Beyond acoustics and sound perception, these techniques can find applications for the modeling of a variety of systems, including computational vision in robotics, haptics and clinical applications.