Recent advances in machine learning have significantly facilitated the study of quantum phases and quantum phase transitions in equilibrium. When applied to dynamical phenomena, both classical and quantum networks demand an extensive cost of resources owing to the inclusion of the temporal dimension. Here, we propose a hybrid quantum‐classical architecture combining a quantum convolutional neural network (QCNN) with a classical neural network (CNN) to study nonequilibrium dynamics, and demonstrate an application of classifying dynamical quantum phase transitions (DQPTs). The QCNN layer of the hybrid network is responsible for extracting spatial features from quantum states at each time step, and the CNN layer processes the temporal sequence of QCNN outputs for classification. For a nearest‐neighbor transverse‐field Ising model, numerical simulations show test accuracy exceeding 95% when identifying DQPTs in quench dynamics, with robustness under single‐ and two‐qubit depolarizing noise. Remarkably, the trained network can be applied without modification to a different Ising model with long–range interaction, showing the potential as a migratable classifier. This hybrid framework circumvents the exponential cost of classical methods and the noise susceptibility of deep quantum circuits, offering a scalable and hardware‐efficient approach for characterizing nonequilibrium dynamics.
Li et al. (2026) studied this question.