Daniel BRUNNER : "Photonic Neural Networks implemented in delay or spatio-temporal networks"

Soutenance d'Habilitation à diriger des recherches : "Photonic Neural Networks implemented in delay or spatio-temporal networks"

Résumé :

Neural Networks (NN) and Artificial Intelligence are revolutionizing computation and having dramatic impact on both science and society. In benchmarks like image recognition or complex decision making, NNs already outperform humans. Although first proposed seventy years ago, NNs have until recently been considered only of minor practical relevance - not for lack of potential, but for the lack of sufficiently powerful computing systems. Even with today’s powerful computers the performance and efficiency remain frustratingly low. The underlying reason is that no concepts to integrate networks have stood the test of time. Using photonics, I will demonstrate the creation of precisely such networks. Crucially, our concepts are scalable to relevant system sizes and ultimately compatible with integration, a fundamental difference to electronics.

In this seminar, I will give a general introduction into information processing, nonlinear systems and Neural Networks. Special attention will be given to resulting fundamental requirements for hardware implementations. In the second part, I will show how we have successfully transferred such computing concepts into photonic hardware. Specifically, I will discuss fundamental advantages and challenges of photonics and illustrate its particular appeal for NN hardware. Finally, I will give an outlook with respect to future developments and the most pressing questions the field needs to address.

Composition du jury :

• Prof. Gordon Pipa, Institute of Cognitive Science, Osnabrück University, Germany. (Rapporteur)
• Prof. Sylvain Gigan, Sorbonne Université, Kastler-Brossel Laboratory, Paris, France. (Rapporteur)
• Prof. Demetri Psaltis, Ecole Polytechnique Fédérale de Lausanne, Lausanne, Switzerland. (Rapporteur)
• Dr. Maria-Pilar Bernal, CNRS, FEMTO-ST, Besançon, France. (Examinateur)
• Prof. Laurent Larger, Université Bourgogne Franche-Comté, FEMTO-ST, Besançon, France. (Examinateur)

Localisation : Amphithéâtre J-J GAGNEPAIN, TEMIS, 15 B Avenue des Montboucons, Besançon