Abstract

Time-Delayed Reservoir computing (TDRC) is an approach to implement artificial neural using a single physical dynamical system or node with a delayed feedback loop. Various photonic implementations have been proposed, most notably those using semiconductor lasers with coherent optical feedback [1,2]. One of the metrics used to assess the performance of reservoir computing is the dynamical consistency [3], which evaluates the system's capacity to generate similar outputs with different initial conditions and when driven with similar input signals. Theoretical and experimental analysis of the impact of operational parameter in laser-based TDRC have been performed with a limited resolution parametric [2]–[4]. They have suggested that the “edge” of the consistency region has a smooth geometric shape [7] and provides enhanced computing capabilities for laser based TDRC.

© 2023 IEEE