Gaussian states provide universal and versatile quantum reservoir computing
Nokkala, Johannes;Martínez-Peña, Rodrigo;Giorgi, Gian Luca; Parigi, Valentina;Soriano, Miguel C.;Zambrini, Roberta
We establish the potential of continuous-variable Gaussian states in performing reservoir computing with linear dynamical systems in classical and quantum regimes. Reservoir computing is a machine learning approach to time series processing. It exploits the computational power, high-dimensional state space and memory of generic complex systems to achieve its goal, giving it considerable engineering freedom compared to conventional computing or recurrent neural networks. We prove that universal reservoir computing can be achieved without nonlinear terms in the Hamiltonian or non-Gaussian resources. We find that encoding the input time series into Gaussian states is both a source and a means to tune the nonlinearity of the overall input-output map. We further show that reservoir computing can in principle be powered by quantum fluctuations, such as squeezed vacuum, instead of classical intense fields. Our results introduce a new research paradigm for quantum reservoir computing and the engineering of Gaussian quantum states, pushing both fields into a new direction.