BEIJING, May 29, 2026 /PRNewswire/ — WiMi Hologram Cloud Inc. (NASDAQ: WiMi) (“WiMi” or the “Company”) is a leading global Hologram Augmented Reality (“AR”) Technology provider. A quantum deep convolutional neural network technology oriented toward image recognition tasks has achieved phased progress. This technology provides a new technical path to address the challenges faced by traditional deep learning in terms of computational complexity, memory consumption, and training efficiency by constructing a quantum deep convolutional neural network model and combining it with a quantum-classical hybrid training mechanism. The proposal of this technology marks a further deepening of the application of quantum machine learning in typical artificial intelligence tasks such as image recognition, and also provides a new research direction for the realization of future large-scale quantum intelligent computing systems.
WiMi has proposed a quantum deep convolutional neural network model for image recognition tasks. The model takes quantum parameterized circuits as its core computing structure, performs feature extraction on image data through quantum convolutional layers, and utilizes a quantum classification layer to complete the final recognition task. The overall architecture draws on the hierarchical structure of classical deep convolutional neural networks in terms of design philosophy, while fully leveraging the parallel computing capability of quantum circuits, enabling the model to achieve higher computational efficiency when processing high-dimensional data.
At the technical architecture level, the quantum deep convolutional neural network consists of a data encoding module, a quantum convolutional layer module, a quantum feature fusion module, and a quantum classification module. The system first maps classical image data to the quantum state space through the data encoding module. Since quantum computers process quantum state information, it is necessary to convert pixel information into probability amplitudes of qubits through specific encoding strategies. This process is usually achieved through amplitude encoding, angle encoding, or hybrid encoding methods, enabling image data to be effectively processed by quantum circuits.
After completing data encoding, the quantum convolutional layer begins to perform feature extraction on the quantum states. Similar to the convolution kernels in classical convolutional neural networks, the quantum convolutional layer operates on local qubits through a set of parameterized quantum gates. These quantum gates can form functions similar to convolution filters, performing feature mapping on the input quantum states. Since quantum gate operations can act on multiple superposition states simultaneously, they enable highly parallel feature extraction processes when handling complex image structures.
The core of the quantum convolutional layer lies in the design of parameterized quantum circuits. WiMi’s circuit consists of basic quantum logic gates such as rotation gates, control gates, and entanglement gates, controlling the evolution process of quantum states through trainable parameters. Rotation gates are used to adjust the state angles of qubits, control gates are used to construct correlation relationships between qubits, and entanglement gates enable the establishment of complex quantum entanglement structures among multiple qubits. Through these operations, the quantum circuit can form feature extraction capabilities similar to those of classical convolutional layers while possessing higher expressive power.
As the quantum convolutional layers are stacked layer by layer, the network is able to gradually extract higher-level image features. Shallow quantum convolutional layers are primarily responsible for capturing low-level features such as edges and textures in the image, while deeper quantum convolutional layers can identify more complex shapes and structural information. Since quantum states remain in superposition during the computation process, the entire feature extraction process can be performed in parallel within an exponentially large state space, thereby significantly improving computational efficiency.
Following the quantum convolutional layers, the system introduces a quantum feature fusion module. This module integrates feature information from different qubits through additional quantum gate operations. Through quantum entanglement mechanisms, image features from different regions can be effectively fused, thereby forming higher-dimensional feature representations with greater discriminative power. Compared to the feature fusion methods in traditional neural networks that rely on matrix multiplication, quantum feature fusion completes information integration through the quantum state evolution process, which can reduce some computational overhead.
After completing feature extraction and fusion, the network enters the quantum classification stage. The quantum classification layer outputs classification results by measuring the probability distribution of the quantum states. Specifically, the quantum circuit measures several key qubits in the final stage and determines the category of the input image through statistical analysis of the measurement probabilities. This process is similar to the fully connected classification layer in classical neural networks, but its computation occurs in the quantum state space, thus enabling the use of quantum parallelism to improve computational efficiency.
To train this quantum deep convolutional neural network, WiMi proposed a quantum-classical hybrid training scheme. Since current quantum hardware is still in the development stage, relying entirely on quantum devices for large-scale training presents certain difficulties. Therefore, the hybrid training strategy has become an effective solution. In this scheme, the quantum circuit is responsible for executing the forward computation process, while parameter updates are completed by classical computers.
During the training process, the system first encodes the image data into quantum states and completes feature extraction and classification calculations through the quantum circuit. Subsequently, by statistically analyzing the quantum measurement results, the error between the network output and the true label is obtained. Classical optimization algorithms calculate gradient information based on this error and update the trainable parameters in the quantum circuit. These parameters are then reloaded into the quantum circuit for the next round of computation, thereby forming a training process in which quantum and classical computing work collaboratively.
WiMi’s hybrid training mechanism draws on the design philosophy of variational quantum algorithms. Variational quantum algorithms combine parameterized quantum circuits with classical optimizers, enabling quantum computing to solve complex problems under limited quantum resources. In this quantum deep convolutional neural network, the idea of variational quantum algorithms is applied to network parameter updates, thereby realizing the feasibility of model training.
From the perspective of computational complexity, the model is theoretically capable of providing significant computational advantages. Traditional deep convolutional neural networks typically exhibit polynomial growth in computational complexity with increasing network scale when processing high-dimensional image data. In contrast, the quantum deep convolutional neural network leverages quantum superposition and parallel computing capabilities to simultaneously process a large number of data states in an exponentially large state space, thereby achieving exponential computational acceleration in certain tasks.
In terms of experimental validation, WiMi conducted quantitative experimental testing of the model on a quantum simulation platform. The experimental results show that the quantum deep convolutional neural network can effectively learn image features in image classification tasks and achieve stable recognition performance. Although the current experimental scale is still limited by the number of qubits, the results have already demonstrated the feasibility of the model in image recognition tasks.
In terms of system implementation, the R&D team has built a complete software and algorithm framework. This framework supports functions such as quantum circuit construction, data encoding, training optimization, and model evaluation, enabling the quantum deep convolutional neural network to run on existing quantum simulation environments and early quantum hardware.
From the perspective of technological development trends, with the continuous advancement of quantum hardware, quantum machine learning models are expected to gradually move toward practical applications. As one of the important directions in quantum machine learning, quantum deep convolutional neural networks have broad application prospects in fields such as image recognition and video analysis. By leveraging the parallel capabilities of quantum computing, such models may demonstrate significant advantages when processing ultra-large-scale data.
WiMi stated that it will continue to optimize the structural design of this quantum deep convolutional neural network in the future, including improving the quantum convolutional layer structure, optimizing quantum data encoding methods, and enhancing the training efficiency of quantum circuits. At the same time, it also plans to explore more complex quantum neural network architectures, such as quantum residual networks, quantum attention mechanisms, and quantum generative models, to further improve the performance of quantum machine learning systems.
In addition, with the integration of quantum computing and high-performance computing platforms, hybrid quantum-classical computing architectures are expected to become an important component of next-generation intelligent computing systems. The research on quantum deep convolutional neural network technology not only provides a new technical route for image recognition but also lays an important foundation for the future development of quantum artificial intelligence.
WiMi’s technology demonstrates the potential value of quantum computing in the field of artificial intelligence. By combining the physical properties of quantum computing with the model structures of deep learning, new intelligent systems with higher expressive power and stronger computational efficiency can be constructed. As quantum hardware gradually matures, such technologies are expected to achieve larger-scale applications in the future. Overall, this quantum deep convolutional neural network technology for image recognition provides new ideas for the development of quantum machine learning. In the future, as quantum computing technology continues to advance, such innovative technologies that integrate quantum computing and artificial intelligence are expected to play an even more important role in the field of intelligent computing.
About WiMi Hologram Cloud
WiMi Hologram Cloud Inc. (NASDAQ: WiMi) focuses on holographic cloud services, primarily concentrating on professional fields such as in-vehicle AR holographic HUD, 3D holographic pulse LiDAR, head-mounted light field holographic devices, holographic semiconductors, holographic cloud software, holographic car navigation, metaverse holographic AR/VR devices, and metaverse holographic cloud software. It covers multiple aspects of holographic AR technologies, including in-vehicle holographic AR technology, 3D holographic pulse LiDAR technology, holographic vision semiconductor technology, holographic software development, holographic AR virtual advertising technology, holographic AR virtual entertainment technology, holographic ARSDK payment, interactive holographic virtual communication, metaverse holographic AR technology, and metaverse virtual cloud services. WiMi is a comprehensive holographic cloud technology solution provider. For more information, please visit http://ir.wimiar.com.
Translation Disclaimer
The original version of this announcement is the officially authorized and only legally binding version. If there are any inconsistencies or differences in meaning between the Chinese translation and the original version, the original version shall prevail. WiMi Hologram Cloud Inc. and related institutions and individuals make no guarantees regarding the translated version and assume no responsibility for any direct or indirect losses caused by translation inaccuracies.
