WiMi (NASDAQ:WIMI) Launches Multi-Channel Quantum Convolutional Neural Network for AI Applications

Highlights

• WiMi launched the MC-QCNN, a quantum convolutional neural network capable of processing multi-channel data for applications in image, video, and medical AI tasks.
• The new quantum convolution kernels use entanglement and gate-level interactions to capture high-order cross-channel correlations beyond classical CNN capabilities.
• A hybrid quantum-classical training system ensures stable learning, while learnable quantum pooling preserves critical feature information across channels.

WiMi Hologram Cloud Inc. (NASDAQ:WiMi) has unveiled a new technology, the Multi-Channel Quantum Convolutional Neural Network (MC-QCNN), designed to enable quantum models to efficiently process multi-channel data. This development positions quantum AI technology for practical applications across industries including image classification, medical imaging, video analysis, and multimodal monitoring, marking a step toward commercial implementation of quantum deep learning.

A New Quantum Neural Architecture

WiMi’s MC-QCNN introduces a hardware-adaptable quantum convolution kernel structure, optimized for multi-channel data. Unlike classical CNNs that rely on sliding pixel-based convolution, the new quantum convolution kernel encodes data into quantum states using amplitudes, phases, and entanglement. Feature fusion occurs through gate-level interactions, allowing the model to capture complex high-order correlations across channels that classical networks cannot replicate.

The system integrates a complete design framework, covering kernel structure, qubit layout, channel encoding, learnable weights, interpretability, and adaptation to hardware constraints. WiMi designed the kernels using single-bit rotation gates, controlled parameterized gates, SWAP interleaving structures, weak entanglement layers, and channel interaction gates, forming a convolution operator capable of representing complex functions while maintaining resistance to quantum decoherence.

Quantum Pooling and Hybrid Training

After convolution, features are downsampled through learnable quantum pooling circuits, designed to preserve key information while reducing quantum state dimensions. This approach addresses the feature loss often seen in conventional QCNNs and improves stability and retention of multi-channel feature information.

Training occurs within a hybrid quantum-classical framework, where classical computing modules handle loss calculation and parameter updates, while the quantum module performs forward propagation and state evolution. Techniques such as extended parameter shift rules, quantum noise simulation, and gradient clipping ensure stability and performance on real quantum hardware.

High-Dimensional Multi-Channel Learning

WiMi observed that the MC-QCNN model automatically identifies nonlinear correlations between channels. For example, when processing RGB images, the convolution kernels do not analyze R, G, and B channels separately. Instead, the kernels establish entanglement-based correlations, recognizing complex patterns across all channels in a high-dimensional quantum state space. This capability allows the network to extract deep features beyond what conventional 3×3 or 1×1 convolutions can achieve.

The company believes that multi-channel processing is essential for quantum neural networks to move beyond academic research and handle real-world data. The MC-QCNN framework positions quantum AI for commercial applications, enabling processing of multi-channel images, video, audio, text, graphs, and sensor data.

Future Development and Industry Impact

WiMi plans to continue refining its quantum neural technology by enhancing kernel efficiency, developing improved noise adaptation strategies, and extending convolution structures to 3D and time-series data. Integration with advanced AI models, including Transformers, is also planned. The company envisions quantum AI becoming a core tool for next-generation general AI applications, with implications for industries ranging from healthcare to multimedia analysis.

Market Activity

WIMI shares last traded at USD 2.75 per share on January 5, 2026.