Mario Chemnitz, Friedrich-Schiller University Jena

Leuchs-Russell Auditorium, A.1.500, Staudtstr. 2

Location details

Abstract:

The current achievements in artificial intelligence are based on decades of progress in the understanding of artificial neural networks as trainable, non-linear learning systems. Decades of knowledge about complex-dynamic systems of physics could now offer a new starting point for intrinsically neuromorphic processors. The lecture sheds light on the basic principles of neural-like computing with nonlinear wave dynamics using current examples and deepens the principle using an experimental demonstration of the principle based on optical waves in nonlinear waveguides. In particular, it is shown that broadband frequency generation in a single fiber make it possible to imitate the classification and prediction functions of different optical networks and even deep nets for some tasks. I’ll further illustrate the mathematical resemblance between nonlinear Schrödinger systems and neural nets and pathways to control optical dynamics to harness this fundamental parallelism. The technological potential of highly dynamic, physical systems as the basis of more energy-efficient computer hardware is up for discussion.