Time series anomaly detection using superconducting transmon quantum computers and error mitigation

We present a low circuit depth Quantum Machine Learning algorithm designed to tackle the widely applicable task of detecting anomalous behavior in time series data. The algorithm follows a one-class-classification approach where anomalous behavior is measured by deviation from a model trained using time series data under “normal conditions”. This is achieved by training a distribution of parameterized Hamiltonians to time-devolve quantum-encoded time series data such that the expectation value of a given observable clusters about a given point; a procedure we refer to as Quantum Variational Rewinding (QVR). To demonstrate the readiness of QVR to tackle real problems on present noisy quantum hardware, we use two IBM superconducting transmon chips to detect anomalies in cryptocurrency trading data resulting from the large movements of bitcoin and USD tether on the blockchain. We also demonstrate how performance can be improved using error mitigation techniques including Pauli twirling and dynamical decoupling.