Recovery of quantum correlations using machine learning

Quantum sources with strong correlations are essential but delicate resources in quantum information science and engineering. Decoherence and loss are the primary factors that degrade nonclassical quantum correlations, with scattering playing a role in both processes. In this work, we present a method that leverages long short-term memory, a machine-learning technique known for its effectiveness in time-series prediction, to mitigate the detrimental impact of scattering in quantum systems. Our setup involves generating two-mode squeezed light via four-wave mixing in warm rubidium vapor, with one mode subjected to a scatterer to disrupt quantum correlations. Mutual information and intensity-difference squeezing between the two modes are used as metrics for quantum correlations. We demonstrate a 74.7% recovery of mutual information and an 87.7% recovery of two-mode squeezing, despite significant photon loss that would otherwise eliminate quantum correlations. This approach marks a significant step toward recovering quantum correlations from random disruptions without the need for hardware modifications, paving the way for practical applications of quantum protocols.