CNN-LSTM neural network for high-fidelity, short-time state measurement and calculation of its cutoff detection time based on Jasen-Shannon divergence

Fast and highly-accurate state measurements play an imperative role in quantum information technology. The state measurement of ion-trap-based qubits is performed by measuring photons from state-dependent scattering. The measurement fidelity is limited by the off-resonant transition during measurement and the imperfection of detectors such as dark counts. To address this problem, methods of utilizing time information of emitted photons have been studied to distinguish actual signals from false signals. It has been reported that the use of machine learning techniques can further improve the measurement fidelity since machine learning techniques can consider complex factors.

In this paper, we propose a combination of convolutional neural network and long short-term memory (CNN-LSTM) model to improve the measurement fidelity. The model is designed to combine the strengths of each model. That is, CNN is specialized in capturing local features, and LSTM is specialized in processing sequential time information. The result shows that the combined model outperforms both CNN and LSTM models and it is robust to the size of time information.

Moreover, we show that the amount of available information from the histograms of each state can be calculated based on Jasen-Shannon divergence. This calculated result can be exploited to determine the offset detection time and discard the redundant information, which in turn can save computational resources.