Time and wavelength multiplexing in photonic neural networks

In this master thesis we explore the reservoir computing paradigm, with a particular focus on delay-based reservoir computing. This concept posits that a large physical network of nodes can be replaced by a single nonlinear node with delayed feedback. In this setup, actual nodes in the physical network become virtual nodes in the delay system via temporal multiplexing. We test the efficiency of a photonic approach in a benchmark task: predicting chaotic time series. We model the system using semiconductor laser-based photonic reservoir computing and time-delayed optical feedback based on the Lang-Kobayashi equations. In addition, we extend this concept by optimising resources, using different frequency filters at the output of the laser, and demonstrate the improvements these filters bring to the prediction task. Our numerical results show that prediction performance is improved when frequency and time multiplexing are combined, suggesting a promising strategy for future experimental implementations.

Supervisors: Miguel C. Soriano, Claudio R. Mirasso

Jury: Pere Colet, Apostolos Argyris, Miguel Cornelles Soriano

Presential, with parallel Zoom stream:  

https://zoom.us/j/98286706234?pwd=bm1JUFVYcTJkaVl1VU55L0FiWDRIUT09