Every quantum system is coupled to an environment. Such system-environment interaction leads to temporal correlation between quantum operations at different times, resulting in non-Markovian noise. In principle, a full characterization of non-Markovian noise requires tomography of a multitime processes matrix, which is both computationally and experimentally demanding. In this paper, we propose a more efficient solution. We employ machine learning models to estimate the amount of non-Markovianity, as quantified by an information-theoretic measure, with tomographically incomplete measurement. We test our model on a quantum optical experiment, and we are able to predict the non-Markovianity measure with $90\%$ accuracy. Our experiment paves the way for efficient detection of non-Markovian noise appearing in large scale quantum computers.

• Received 4 February 2021
• Revised 17 June 2021
• Accepted 4 August 2021

DOI:https://doi.org/10.1103/PhysRevA.104.022432