Reservoir computing using a delayed feedback system: towards photonic implementations

Delayed feedback systems are known to exhibit a rich dynamical behavior, showing a wide variety of dynamical regimes. We use this richness to implement reservoir computing, a processing concept in machine learning. In this paper we demonstrate the proof of principle on an electronic system, however the approach is readily transferable to photonics, promising fast and computationally efficient all-optical processing. Using only one single node with delayed feedback instead of an entire network of nodes, we succeed in obtaining state-of-the-art results on benchmarks such as spoken digit recognition and system identification.