Multicolor Tunable Photonic Reservoir Computing

Abstract

Reservoir computing (RC) is a bioinspired computational paradigm that employs nonlinear dynamical systems (i.e. reservoir) to increase the dimensionality of sequential data [1] . The reservoir is fixed and only the readout is trained with a simple method such as linear regression to map the states of the reservoir to a targeted pattern. A delay line together with a nonlinear node, constitutes the basic topology of a delay-feedback reservoir. The complex dynamics of such systems is engaged with time multiplexing to create a number of virtual time nodes over the delay time associated with the feedback loop. The delay time is usually harmonized to the time sequence of the input data. To create a sufficiently strong nonlinear mapping, the independent internal states of the reservoir must be increased by time multiplexing at a rate much faster than the delay time. The number of virtual nodes created by time multiplexing is thus limited by the speed at which the input data is masked (e.g. by modulating the sampled-and-hold input data). In most applications, hundreds of nodes are typically required. This necessitates multiplexing with the speeds hundreds of times faster than the input data stream. To overcome the inherent trade-off between the number of neurons and the processing time, parallelization schemes might be envisioned [2] , [3] .