Kiryu Shohei

We examine photonic how to implement quantum reservoir computing with stochastic internal states on existing hardware, using a classical simulator imitating Xanadu’s photonic quantum computer, Aurora. The proposed model extracts reservoir features from measurement probabilities and modifies Aurora’s native circuit by introducing probabilistic routing between squeezed coherent and squeezed vacuum states. The framework is supported by the universality of the stochastic reservoir map, which can approximate polynomial and more general continuous functions on compact domains. In benchmark simulations, for MNIST classification with reservoir size 1200, the quantum reservoir achieved 91.4% accuracy, slightly outperforming a same-size classical echo state network. The results also suggest that readout stochasticity can improve generalization by acting as an effective regularizer.