Correcting measurement errors in multiqubit circuits

Reducing measurement errors is critical for performing any quantum algorithm, especially those requiring feedback. In certain situations, e.g. in quantum experiments that only require averaged measurements, errors can be corrected using post-processing techniques. These experiments include any that only require operator expectation value measurements, such as tomography or variational quantum eigensolvers. Here we will present one such method, which involves calibrating these errors via measurements of prepared computational states which are used to feed a neural network. In this way we reduce the experimental overhead from exponential to polynomial scaling. We will present experimental demonstrations of this technique on IBM Q devices.