WiMi's Next-Generation Quantum Convolutional Neural Network Reshapes Classical Data Classification Methods

WiMi Hologram Cloud Inc. (NASDAQ: WiMi ) ("WiMi" or the "Company"), a leading global Hologram Augmented Reality ("AR") Technology provider, proposes a cutting-edge quantum machine learning technology oriented toward classical data classification tasks—a quantum convolutional neural network with interaction layers for classical data classification. This technology systematically enhances the overall performance of quantum convolutional neural networks in terms of expressive power, entanglement generation capability, and actual classification performance by introducing a novel interaction layer structure based on three-qubit interactions, marking an important step forward in the structural design of quantum deep learning models toward a new phase driven by multi-body interactions. From the perspective of technical implementation logic, this quantum convolutional network adopts an overall hybrid quantum-classical architecture design. First, classical data is mapped to the quantum state space through an efficient data encoding strategy, ensuring that as much discriminative information from the original data as possible is preserved under limited qubit resources. For image data, the network employs block partitioning and local mapping approaches to embed pixel information into corresponding quantum subsystems; for one-dimensional data, a combination of structured amplitude encoding and angle encoding is used to achieve a compact representation of data features. After data encoding is completed, the quantum state is fed into the quantum feature extraction module composed of multiple layers of quantum convolutional units and interaction layers. In this module, quantum convolution operations and the novel interaction layers are executed alternately. The quantum convolutional layers are responsible for extracting low-order features within local qubit subspaces, with their structural design adhering to hardware-friendly principles to avoid introducing excessively deep or difficult-to-implement quantum gate sequences. The interaction layers serve as the key innovation of the entire network, achieving cross-channel and cross-scale information fusion through three-qubit interactions. This design enables the network to significantly enhance its expressive power for complex patterns while keeping circuit depth under control. The WiMi R&D team systematically studied the impact of this interaction layer on the coverage capability of the quantum state space in theoretical analysis. The results show that after introducing three-body interactions, the set of reachable states in the parameter space of the network is significantly expanded, effectively alleviating the common expressivity limitation problem in traditional quantum neural networks. In terms of entanglement capability, WiMi further conducted an in-depth analysis of the proposed network structure from the perspective of quantum information theory. The study shows that the three-qubit interaction layer can generate high-intensity, multi-scale entanglement structures at relatively shallow circuit depths, which is crucial for quantum machine learning models to capture nonlinear correlations in the input data. Compared to network structures that rely solely on two-qubit entanglement gates, the new model exhibits clear advantages across multiple metrics, including entanglement entropy, uniformity of entanglement distribution, and efficiency of entanglement propagation. This characteristic not only enhances the model's learning capability but also provides strong support for maintaining stable performance under the presence of noise. In terms of the training mechanism, this quantum convolutional neural network employs a joint iterative approach between classical optimizers and quantum circuit parameters to complete model learning. The output of the quantum circuit is mapped into classical feature vectors through measurement, which are then evaluated by.