BEIJING, July 20, 2026 /PRNewswire/ -- WiMi Hologram Cloud Inc. (NASDAQ: WiMi) ("WiMi" or the "Company"), a leading global Hologram Augmented Reality ("AR") Technology provider, announced the release of its latest independently developed achievement—a quantum convolutional neural network for efficient image classification based on quantum random access memory (QRAM). This technology is oriented toward typical image classification tasks with large-scale input data and multiple output channels. It systematically addresses the key bottlenecks of existing quantum convolutional neural networks in terms of the number of qubits, circuit depth, and data loading efficiency, providing a feasible path for the practical application of quantum computing in real machine learning scenarios.

The success of traditional convolutional neural networks on classical computers relies on highly parallel matrix operations and massive storage resources. However, as input image resolution continues to increase and the number of feature channels keeps growing, the resource demands of CNNs in both training and inference stages grow exponentially. This not only increases the hardware burden but also limits the model's application space in edge computing, low-power devices, and high-real-time scenarios. The proposal of quantum convolutional neural networks is precisely an attempt to significantly compress the required computational resources while maintaining model expressiveness through quantum superposition and quantum parallelism.

However, existing QCNN models still face severe challenges at the engineering level. On one hand, the number of qubits in currently available quantum computing devices is limited and noise levels are high, making it difficult to directly handle large-scale input data; on the other hand, many quantum neural network schemes use amplitude encoding or angle encoding in the data encoding stage, requiring individual loading of each input sample—this process itself consumes a large amount of circuit depth, offsetting the potential advantages of quantum computing. How to process large-scale image data in a one-time, efficient manner under limited quantum resources has become the core problem constraining the practicalization of QCNNs.

In response to the above issues, WiMi re-examined the data access problem in quantum neural networks from the architectural level and introduced quantum random access memory (QRAM) as a key technology into the overall design of quantum convolutional neural networks. The core idea of QRAM is to use quantum superposition states to access multiple memory addresses simultaneously, enabling massive classical data to be indexed and invoked using a logarithmic number of qubits. This characteristic makes it naturally suitable for combination with the large-scale feature representation needs in deep learning.

In this released technical scheme, QRAM is not merely used as a simple data loading tool but is deeply embedded into the feature extraction and channel mapping process of QCNN, forming an entirely new model structure. After the input image is mapped into a data structure suitable for quantum storage during the classical preprocessing stage, the corresponding quantum state representation is constructed through QRAM. Unlike traditional pixel-by-pixel or block-by-block loading methods, this scheme allows multiple spatial locations and feature dimensions to exist simultaneously in the quantum state, thereby enabling parallel access to large-scale input data in a single quantum operation.

At the convolution computation level, the model proposed by WiMi redefines the implementation of quantum convolution kernels. In traditional QCNN, convolution operations are often realized through local quantum gate combinations, which significantly increase circuit depth as input scale grows. This technology, leveraging the parallel addressing capability provided by QRAM, matches convolution kernel parameters and input features in the form of quantum states, transforming the convolution process into a series of controlled quantum operations. This design significantly weakens the coupling relationship between the depth growth of the convolution layer and input size, allowing the model to maintain a relatively shallow quantum circuit structure even when processing high-resolution images or multi-channel features.

In terms of output channel expansion, this technology also embodies the design philosophy of resource efficiency. In classical CNNs, increasing the number of output channels often leads to linear or even super-linear growth in computation and storage demands, while in quantum architectures, blindly expanding channels would quickly exhaust available qubit resources. The channel mapping mechanism based on QRAM proposed by WiMi introduces auxiliary index registers in the quantum state, allowing multiple output channels to exist in superposition within the same quantum circuit, with selective readout during measurement to complete classification decisions. This mechanism significantly reduces the direct consumption of quantum resources caused by channel count growth, enabling the model to adapt to more complex classification tasks.

In the overall training and inference process, this QCNN model adopts a typical hybrid quantum-classical architecture. Parameter updates and loss function evaluation are completed on the classical computing side, while core feature extraction and mapping processes are handled by quantum circuits. This design fully considers the current development stage of quantum hardware, avoiding dependence on large-scale fault-tolerant quantum computers while maximizing the advantages of quantum computing in high-dimensional feature space processing. Through repeated iterative optimization, the model achieves a good balance among qubit count, circuit depth, and runtime while ensuring classification accuracy.

During the experimental validation phase, WiMi conducted systematic evaluations of this model on multiple sets of image classification tasks with different scales. The results show that, under conditions of significantly increased input data scale and continuously expanded output channel count, the quantum convolutional neural network based on QRAM outperforms existing similar QCNN schemes in both resource consumption and circuit depth while maintaining competitive classification performance. This result validates the feasibility of this technology in processing large-scale machine learning tasks in resource-constrained quantum environments.

WiMi's quantum convolutional neural network for efficient image classification based on quantum random access memory (QRAM) not only expands the design space of quantum machine learning models in theory but also provides a practically feasible solution path for large-scale data processing at the engineering level. With the continuous improvement of quantum hardware capabilities, the QRAM-based QCNN architecture is expected to further amplify its parallel computing advantages, offering new solutions for complex visual tasks and laying a solid foundation for the actual deployment of future quantum intelligent systems.

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.