Machine Learning of Noise in Single-Qubit Hardware

As quantum information processors (QIPs) grow from 2, to 5, to 16 or more qubits, characterizing their behavior rapidly becomes challenging. Techniques commonly used today, such as tomography and randomized benchmarking, are unlikely to scale easily to many qubits while providing useful debugging information. QIP development will require fast, scalable, and accurate techniques that extract useful information about noise affecting QIPs and the errors they are likely to suffer in use. Machine learning tools are a promising alternative to the brute force and/or ad-hoc statistical methods that underlie most existing techniques. Here, we demonstrate a machine learning classifier that distinguishes whether the noise on a single-qubit QIP is stochastic or coherent. The classifier uses data from certain structured circuits, specifically those used for gate set tomography, but does not rely on any of the standard statistical tools for analyzing such data, and can in principle be applied to arbitrary data that contains information about the property of interest.