Design of quantum dot networks for improving prediction performance in reservoir computing

A quantum dot (QD) network generates various fluorescence signals based on nonlinear energy dynamics which depend on its structure and composition and is utilized for a component of physical reservoir computing. However, existing designs rely on random QD networks, which is not be optimal for enhancing the prediction performance. In this paper, we propose a method for designing effective quantum dot (QD) networks to improve the performance of reservoir computing. The fluorescence signals from numerous virtual QD networks can be reproduced through numerical simulation based on a deterministic mathematical model, and the QD networks generating the most significant signals contributing to the prediction performance are identified. We demonstrated that QD reservoir computing using designed QD networks predicts time-series data more accurately than using random QD networks in the numerical simulations.