Influence of measurements on Continuous Variable Quantum Reservoir Computing

Abstract

[eng] Reservoir Computing is a paradigm of Machine Learning that harnesses the information processing potential of complex dynamical systems. Recently, quantum systems have been suggested as promising candidates for reservoir computing due to the significant growth in degrees of freedom they allow against their classical counterparts. In this thesis, a photonic platform is proposed for reservoir computing and on-line data processing, in the realm of continuous variable quantum optics. A non-linear medium provides the needed interaction between different optical modes. The output light is splitted in a re-injected and a measured component. Control of the needed memory is effectively achieved through the coherent optical feedback. The problem of quantum measurements and back-action are addressed for the first time in this context. The performance of the model is numerically tested for each of the main parameters that can be externally tuned. The role of quantum effects, i.e. squeezing, is also addressed in connection with the computational capacity of the system. To conclude, aspects relevant for possible experimental implementations are commented