WiMi Studies Hybrid Quantum-Classical Inception Neural Network Model for Image Classification

BEIJING, Feb. 18, 2026 /PRNewswire/ -- WiMi Hologram Cloud Inc. (NASDAQ:WIMI) ("WiMi" or the "Company"), a leading global Hologram Augmented Reality ("AR") Technology provider, proposed a brand-new technological innovation—a hybrid quantum-classical Inception neural network model for image classification. This is a brand-new hybrid architecture that naturally integrates quantum computing with classical deep learning through Inception-style parallel feature channels, achieving triple improvements in performance, efficiency, and robustness. The core goal of this technology is to utilize the high-dimensional feature expression capability of quantum computing to solve the expressiveness bottleneck of image classification models, while enhancing engineering implementability through classical network structures, and building a research path regarding the relationship between quantum expressiveness, quantum entanglement degree, and model performance, laying the foundation for future hybrid quantum AI research.

Past quantum neural network research has mostly focused on constructing some kind of variational quantum circuit and attempting to embed it into traditional neural network structures. Although this method can achieve improvements in small-scale tasks, overall performance growth is slow and has not fully tapped the potential of quantum computing. For this reason, the WiMi research team realized that to allow quantum computing to play a true role in image classification, its parallel structure must be redesigned, especially needing to break through the structural limitations of single-path quantum networks.

The core idea of the Inception structure is to allow multiple sub-networks with different receptive fields and expression methods to extract features in parallel and then complete multi-scale fusion through concatenation. By re-examining the quantum-classical hybrid network through this idea, WiMi proposed three parallel feature paths:

Quantum feature extraction path: utilizing the multi-dimensional Hilbert space of quantum circuits to perform quantum encoding on local regions of images, and then extracting complex features through parameterized quantum gates and entanglement structures.

Classical feature extraction path: using efficient convolution and lightweight feature extraction units to enhance model stability and macroscopic structure recognition capabilities.

Hybrid quantum-classical path: taking the output of classical convolutional layers as input to quantum circuits, allowing classical features to be mapped into quantum space to obtain higher-order nonlinear expressive capabilities.

The three paths together constitute the parallel Inception module, and then the outputs are concatenated into the final feature tensor to enter the subsequent classifier.

Such a design not only enables the model to simultaneously possess three major advantages—quantum high-dimensional expression, classical strong stability, and cross-domain feature fusion—but also thoroughly solves the industry pain point of training difficulties caused by excessively deep circuits in pure quantum networks. The quantum part does not need to construct extremely deep circuits but instead achieves more expressive space in shallow layers through parallel structures, fundamentally improving model trainability and scalability.

The key to building a hybrid quantum-classical Inception network lies in how to effectively map image data to quantum circuits. WiMi adopted an encoding strategy based on parameterized rotation gates, mapping image blocks to multi-qubit rotation angles so that they can represent complete local features in the quantum state space. Subsequently, the team designed controlled rotation gates, entanglement structures, and depth-adapted quantum circuits to enable quantum states to achieve the highest possible expressiveness in limited depth.

The design of the quantum path follows the principles of shallow circuits, high entanglement, and strong expression. By introducing multiple sets of entanglement constructions, quantum states can rapidly diffuse between different layers and generate higher-order feature combinations. The structural selection of quantum circuits is no longer based on manual guessing but on systematic research into the relationships between expressiveness, entanglement degree, and training stability, thereby constructing the circuit topology most suitable for image classification.

The classical path adopts lightweight convolutional networks to maintain good generalization ability and training efficiency. In the hybrid path, WiMi embeds features extracted by classical convolutions into new quantum circuits for secondary enhancement, enabling the model to possess the capability of first classical understanding followed by quantum enhancement.

The entire Inception module provides the classifier with a richer, more three-dimensional, multi-scale feature expression space by concatenating and fusing features from the three paths. Among them, quantum features serve as high-order expression supplements, classical features are responsible for stable backbone expression, and the hybrid path acts as a bridge to naturally fuse the two.

Through extensive experimental validation, the WiMi research team discovered that the hybrid quantum-classical Inception structure has multiple outstanding advantages. The quantum path can capture highly complex texture variations and subtle patterns in images, the classical path ensures overall stability and robustness, and the hybrid path enables the model to possess cross-domain expressive capabilities. When combined, the model's performance in image classification tasks surpasses that of ordinary convolutional networks and single-path quantum networks, with particularly significant performance in scenarios with small data scales and subtle category differences. In addition, the high-dimensional nature of quantum circuits allows the model to achieve strong expressive power with fewer parameters, thereby realizing the dual advantages of high performance + low parameter count.

WiMi's hybrid quantum-classical Inception neural network is not merely a structural innovation; it represents a future trend: quantum computing will no longer exist as an independent model but will gradually become one of the foundational capabilities of deep learning. By deeply fusing quantum circuits with classical networks in terms of feature domains, information flows, and spatial structures, this model demonstrates a possible operation mode for future intelligent perception systems—quantum and classical parallel collaboration, processing features of different levels and natures in the most appropriate way. In the future, WiMi will continue to explore deeper hybrid structures, more complex quantum feature encoding methods, and deployment methods oriented toward real quantum hardware, promoting hybrid quantum artificial intelligence toward practical applications.

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.

View original content:https://www.prnewswire.com/news-releases/wimi-studies-hybrid-quantum-classical-inception-neural-network-model-for-image-classification-302691534.html

SOURCE WiMi Hologram Cloud Inc.