BEIJING, May 6, 2026 /PRNewswire/ -- WiMi Hologram Cloud Inc. (NASDAQ: WiMi) ("WiMi" or the "Company"), a leading global Hologram Augmented Reality ("AR") Technology provider, launched a breakthrough technological achievement—a Multi-Scale Fusion Quantum Deep Convolutional Neural Network for Text Classification. This technology is based on an advanced quantum convolutional architecture and an innovative multi-scale feature fusion mechanism, aimed at solving bottlenecks in the field of natural language processing (NLP) such as high model complexity, limited embedding representations, and difficulty in scaling quantum networks.

This brand-new Quantum Deep Convolutional Neural Network (QDCNN) not only achieves breakthrough improvements in key dimensions such as parameter scale, computational complexity, and training efficiency but also realizes unified modeling of word-level and sentence-level information, obtaining performance superior to existing state-of-the-art quantum models on multiple standard datasets. The release of this technology is regarded as an important milestone in promoting quantum natural language processing from theory to practical application.

In recent years, with the widespread success of large neural network architectures such as Transformer in the NLP field, the industry's demand for fast, scalable, and low-energy-consumption natural language processing models has been growing. Quantum machine learning is considered a potential key technological direction to break through this dilemma. After in-depth analysis of these bottlenecks, the WiMi R&D team realized that a more structured, scalable network framework capable of fully leveraging the advantages of quantum computing is essential. Therefore, they focused their attention on convolutional structures—a neural network architecture that has been the most time-tested and scalable in human modeling of visual and textual data.

The first key innovation proposed by WiMi is Quantum Depthwise Separable Convolution.

Traditional depthwise separable convolution consists of two steps:

• Depthwise Convolution performed independently on different channels;
• Pointwise Convolution that linearly combines multiple channels.

This structure significantly reduces the number of parameters in classical neural networks, making it the core idea of lightweight CNN design. The WiMi research team cleverly mapped this concept to the quantum circuit architecture, achieving multiple breakthroughs:

First, by encoding input features into quantum states and applying quantum convolution operations in a per-channel manner, the model avoids the exponential increase in parameter consumption that occurs when traditional quantum networks expand in width or depth. Quantum depthwise convolution allows each qubit or qubit cluster to independently process word-level local semantics, thereby preserving the locality advantage of convolution operations.

Second, in the pointwise quantum convolution module, trainable quantum gate combinations are used to achieve interaction and channel fusion between quantum states, successfully compressing multi-dimensional representations into a more expressive semantic space. Quantum gates can achieve higher expressive power than classical linear layers, particularly when operating in high-dimensional Hilbert space, where their modeling capability has a natural advantage.

Ultimately, this quantum depthwise separable convolution significantly reduces the number of controlled rotation gates required by traditional quantum convolution structures, making the model's execution efficiency on simulators and real quantum hardware several times higher than existing quantum convolution models. Quantum depthwise separable convolution not only brings a lightweight structure to quantum NLP models but also solves the scalability problem of quantum networks in text processing, making quantum convolution a core building block for NLP QNNs.

In addition, text classification is a task that relies on both local information and overall semantics. For example, sentiment analysis needs to focus on the polarity of individual words as well as understand the contextual semantics of the entire sentence. To this end, traditional NLP architectures often adopt multi-layer convolution, bidirectional LSTM, or Transformer self-attention structures to simultaneously capture features at different semantic scales.

In quantum NLP research, how to enable quantum models to simultaneously understand local word meanings and global sentence meanings has been a problem that has not yet been fully solved. The multi-scale feature fusion mechanism (Multi-Scale Fusion Mechanism) proposed by WiMi this time effectively fills this gap.

This mechanism consists of two key parts.

The first part is word-level feature extraction, utilizing quantum convolution to extract local n-gram representations, such as sentiment polarity words, adjective structures, negation word combinations, etc. Quantum states can simultaneously encode multiple word combination patterns in a superposition manner, thus having a natural advantage in n-gram modeling.

The second part is sentence-level feature extraction, extracting semantic structures across sentence levels through multi-layer quantum convolution and quantum pooling operations. Quantum pooling achieves dimension compression through measurement and incomplete observation mechanisms while preserving key information in the quantum state, enabling the model to effectively capture the overall theme and paragraph structure of sentences.

Most critically, the proposed feature fusion module can merge word-level and sentence-level features into the same semantic space in a trainable quantum gate manner. The fused quantum state possesses both local sensitivity and the ability to reflect overall semantics, exhibiting richer feature capabilities than traditional QNNs. Through ablation experiments, WiMi found that this multi-scale feature fusion mechanism contributes significantly to the model's final performance improvement, bringing more than 6% accuracy gains on multiple datasets.

On two public text classification benchmarks, the research team conducted complete experimental validation. The results show that this quantum deep convolutional model achieves leading performance on multiple metrics: it improves accuracy while reducing model parameters by more than 30% compared to classical CNNs; it outperforms various existing quantum models, including QRNN, QSAM, and QTF (Quantum Transformer), by 4% to 10% in accuracy; it maintains stable performance even in noisy quantum hardware simulation environments, demonstrating strong noise resistance.

Ablation experiments further validate the practical value of multi-scale feature fusion and quantum depthwise separable convolution, proving that each design element of this architecture makes a key contribution to overall performance.

The launch of this technology by WiMi has profound industrial significance. As quantum computing enters the early practicalization stage, the scale, depth, and adaptability of quantum models will become key competitive points. It is not only a technological innovation but also a profound breakthrough in the field of quantum natural language processing. It validates the feasibility, effectiveness, and scalability of quantum convolutional structures in text processing, bringing a new paradigm to quantum NLP architectures.

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.