An international team led by researchers of the Institute of Cross-disciplinary Physics and Complex Systems (IFISC, UIB-CSIC) has proposed the use of Gaussian states to perform reservoir computing both in quantum and classical regimes. The paper is published in Communications Physics.
The paper demonstrates that Quantum Reservoir Computing operating with Gaussian resources of continuous variables systems is universal for time series processing, even with quantum (vacuum) fluctuations and even at room temperature. Until now, the implementation of Reservoir Computing has been explored only with discrete variables (such as spins), so generalizing its implementation also with continuous variables opens up the possibility of experimental applications based in photonics in the future. They also showed that these new Gaussian resources are sufficient to achieve universality, which implies that it is possible to solve with them any problem that was previously possible to solve by Reservoir Computing.
The discovery that it is sufficient to train only the connections leading to the final output layer in recurrent neural networks without any apparent loss of computational power has fostered the field of Reservoir Computing. It achieves its purpose by using the computational capacity, high-dimensional state space, and memory of generic complex systems, giving it more engineering freedom than traditional computing or recurrent neural networks. Quantum Reservoir Computing is an emergent field among the Machine Learning techniques and has applications in artificial intelligence and time data analysis, such as chaotic time series or channel equalization.
The model proposed in the study can be adapted to many classical and quantum platforms modeled by continuous variables in linear networks. Researchers state that this work reveals the potential of quantum optical systems for Quantum Reservoir Computing.
Gaussian states of continuous-variable quantum systems provide universal and versatile reservoir computing. Nokkala, Johannes; Martínez-Peña, Rodrigo; Giorgi, Gian Luca; Parigi, Valentina; Soriano, Miguel C. and Zambrini, Roberta. Communications Physics. DOI: 10.1038/s42005-021-00556-w