Demonstration of quantum dot reservoir computing based on spatio-temporal optical processing

A quantum dot network, which consists of coupled structures of randomly dispersed quantum dots, has been studied as a nano-scale optical reservoir for effective machine learning processing. In this study, we defined spatio-temporal fluorescence of a quantum dot network as the reservoir output, which is due to the characteristic dynamics of the excited energy in the network induced by laser pulse irradiation. In order to verify whether a quantum dot reservoir can improve the processing efficiency of advanced machine learning applications, we performed experimental reservoir computing using a numerical model. Several parameters that were required for the construction of the model were defined from the spatio-temporal fluorescence of an experimental quantum dot reservoir. Subsequently, the corresponding reservoir computing based on the model was numerically demonstrated. Reliable performances were successfully demonstrated as sufficient error rates toward the delayed XOR task. Additionally, the dependency on quantum dot compositions of these performances was clarified.