WiMi Releases Next-Generation Quantum Convolutional Neural Network Technology for Multi-Channel Supervised Learning

Beijing, Jan. 05, 2026 (GLOBE NEWSWIRE) -- WiMi Releases Next-Generation Quantum Convolutional Neural Network Technology for Multi-Channel Supervised Learning

BEIJING, Jan. 05, 2026––WiMi Hologram Cloud Inc. (NASDAQ: WiMi) ("WiMi" or the "Company"), a leading global Hologram Augmented Reality ("AR") Technology provider, announced the launch of their independently developed new technology: a Quantum Convolutional Neural Network for Multi-Channel Supervised Learning (MC-QCNN). This breakthrough method, for the first time, constructs a fully hardware-adaptable quantum convolution kernel design, enabling quantum models to efficiently process multi-channel data, thereby demonstrating absolute advantages in industries such as image classification, medical imaging, video analysis, and multimodal monitoring.
From a research and development perspective, the core of this technological breakthrough lies not merely in the construction of multi-channel quantum convolution kernels but in the entire systematized design scheme, including convolution kernel structure, qubit layout, channel interaction encoding, weight learnability, interpretability, and hardware constraint adaptation strategies. To enable the technology to be executed on real hardware, WiMi abandoned a large number of impractical deep circuit structures and instead turned to a design philosophy that is closer to the native gate operation characteristics of quantum hardware. The quantum circuit convolution kernel proposed by WiMi uses single-bit rotation gates, controlled parameterized gates, SWAP interleaving structures, weak entanglement layers, and channel interaction gates, thereby forming a convolution operator that can express complex functions while maintaining robustness against quantum decoherence.
Unlike classical convolution kernels that need to slide within pixel neighborhoods, WiMi adopted a quantum-specific encoding method to compress and encode data from multiple channels into the amplitudes, phases, or entanglement structures of quantum states, performing convolution-like processing on them through parameterized quantum gates. Feature fusion between channels no longer relies on linear weighting but directly generates high-dimensional correlations in the quantum state space through gate-level interactions, producing stronger feature combination capabilities than classical convolution. Through training, these parameterized quantum convolution kernels can learn high-order cross-channel features, such as texture-color co-occurrence, time-space joint patterns, multispectral energy distribution correlations, etc., thereby achieving expressive capabilities superior to traditional QCNN.
One of the cores of this technology architecture is the quantum multi-channel convolution operator established by WiMi. This operator uses parameterized rotation gates and controlled gates to construct convolution patterns. By adjusting the rotation angles of the gates and the controlled structures, the convolution kernel can automatically learn the optimal cross-channel feature combination strategy during training. The entire convolution kernel can not only act on single-bit distributions but also act on multi-bit channel structures in a tensor-like manner, enabling the convolution kernel not only to extract local coherence but also to mine high-order relationships from entanglement structures. This mode cannot be directly realized in classical CNNs because the combination of multi-channel features in classical neural networks is usually based on linear superposition, whereas quantum convolution kernels are based on quantum superposition and quantum entanglement, capable of expressing complex multi-channel correlations in an exponential feature space.
After the convolution operation is completed, the feature maps are compressed into more compact quantum states in the quantum system and downsampled by quantum pooling circuits. The pooling circuits have also been redesigned to handle quantum state features from multiple channels. WiMi adopts a learnable quantum pooling mode, reducing quantum state dimensions through controllable measurements or controllable compression operations while preserving key feature information, which avoids the feature destruction problem caused by direct measurements in traditional QCNNs. Experimental results show that the new pooling structure is more stable than traditional QCNN pooling methods and has a higher feature retention rate.
In addition to convolution kernels and pooling circuits, WiMi has also constructed a dedicated hybrid quantum-classical training framework. During the training process, the classical computing module is responsible for loss function calculation, gradient solving, and parameter updating, while the quantum module is responsible for forward propagation and quantum state evolution. WiMi adopts an extended parameter shift rule approach, enabling all parameters in the multi-channel quantum convolution kernel to be effectively trained. To improve training stability, WiMi also introduces quantum noise simulation and gradient clipping mechanisms, ensuring that the model's performance on real quantum hardware does not sharply decline due to noise.
During the training process, the WiMi team observed a highly valuable phenomenon: the model is able to automatically capture nonlinear correlations between multiple channels. Taking RGB images as an example, the quantum convolution kernels learned by the model do not simply perform linear traversal on the R, G, and B channels but instead establish correlations between channels through entanglement layers, enabling the convolution kernel to recognize joint features of color distribution patterns in the quantum state space. This means that the model is not performing convolution separately on the three channels but is learning an overall deep feature in a higher-dimensional space, with expressive power far superior to that of 3×3 or 1×1 convolutions in classical CNNs.
WiMi believes that multi-channel processing capability will become one of the key abilities for quantum neural networks to move toward real-world applications. Although single-channel QCNN has exploratory significance in academia, its limitations make it unable to meet the industry's requirements for complex data. The emergence of MC-QCNN enables quantum deep learning systems to possess the ability to process real-world data for the first time, meaning that quantum AI is no longer just a laboratory concept but is beginning to have the possibility of commercial implementation. It is believed that, with the improvement of quantum hardware performance, this technology will drive quantum machine learning from laboratory research toward a true era of applications.
In the future, WiMi will continue to refine this technology system, including building more efficient quantum convolution kernel structures, developing more robust noise adaptation strategies, extending to three-dimensional convolution and time-series convolution structures, and exploring integration possibilities with model structures such as Transformer, enabling quantum models to process not only multi-channel images but also multimodal speech, video, text, graph structures, and sensor data. Quantum deep learning will no longer be limited to small-scale tasks but will become an important operator in next-generation general AI models. The combination of quantum computing and artificial intelligence will be the core trend in technological development over the next decade. WiMi will continue to dedicate itself to promoting the construction of the quantum AI ecosystem, allowing quantum technology to truly serve industrial needs, social value, and the human future.

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.

Investor Inquiries, please contact:

WIMI Hologram Cloud Inc.
Email: [email protected]

ICR, LLC
Robin Yang
Tel: +1 (646) 975-9495
Email: [email protected]