Improving Readout Fidelity of Circuit QED Quantum Processors using Recurrent Neural Networks

Achieving high-fidelity state readout of quantum processors is vital for implementing efficient quantum algorithms. Typical projective measurement readout protocols in circuit QED use Gaussian Mixture Models with thresholding to classify qubit states, but neglect correlations in time. Prior works have trained traditional machine learning classifiers, such as support vector machines, on time-resolved data to improve state readout fidelity. Motivated by their expressive power in learning long-time correlations in sequential data, we train Long Short-Term Memory (LSTM) recurrent neural network (RNN) classifiers on time-resolved dispersive readout data from a circuit QED quantum processor. We then explore the capabilities of such LSTM RNNs for learning multi-qubit correlations.