High-resolution consistency analysis for performance evaluation of photonic time-delay reservoir computers
Résumé
The artificial neural network architecture known as "reservoir computing" (RC) is a recent machine learning paradigm. The implementation of a reservoir computer on various hardware platforms including photonics is very promising for data processing. We study here an architecture based on a laser diode with time delayed optical feedback. We conduct an in-depth high-resolution study of both the metrics linked to the RC performances and the system dynamic consistency.We show that the boundary of the consistency region is not continuous but has a periodic tongue topology. This structure is only seen when performing a high-finesse scanning of the system parameters, and strongly depend on the input data modulation parameters. Consistency measure also provides insigths into the dynamic of the system under study. Linking it to IA metrics establishes a deep understanding of the operation of a laser diode-based RC.
Origine | Fichiers produits par l'(les) auteur(s) |
---|