Beijing, Feb. 06, 2026 (GLOBE NEWSWIRE) -- WiMi Releases Hybrid Quantum-Classical Neural Network (H-QNN) Technology for Efficient MNIST Binary Image Classification BEIJING, Feb.06, 2026––WiMi Hologram Cloud Inc. (NASDAQ: WiMi) ("WiMi" or the "Company"), a leading global Hologram Augmented Reality ("AR") Technology provider, today announced the release of a Hybrid Quantum-Classical Neural Network (Hybrid Quantum-Classical Neural Network, H-QNN) technology for efficient MNIST binary image classification. This ...

Core Idea - WiMi Hologram Cloud Inc. has introduced a new Lean Classical-Quantum Hybrid Neural Network (LCQHNN) framework designed to enhance learning efficiency while minimizing quantum circuit complexity, marking a significant advancement in the practical application of quantum neural networks [1][11]. Technical Overview - The LCQHNN framework focuses on quantum feature amplification combined with classical stability optimization, creating an efficient interaction mechanism between classical and quantum computing [2]. - The architecture consists of a Classical Front-End for feature extraction and a Quantum Back-End utilizing variational quantum circuits for classification [2]. - The classical component employs lightweight convolutional layers for data preprocessing, embedding results into quantum state space for feature transformation [3]. - The quantum section features a four-layer variational quantum circuit (4-layer VQC) that optimizes circuit parameters to enhance classification performance while reducing resource consumption [4]. Workflow Stages - The workflow includes several key stages: 1. Data Preprocessing and Classical Encoding, where images are processed into medium-dimensional vectors for quantum encoding [5]. 2. Quantum State Preparation and Entanglement Structure Construction, enhancing correlations between qubits to improve model performance [6]. 3. Parameterized Quantum Evolution and Measurable Readout, utilizing adjustable parameters for efficient training and measurement [7]. 4. Classical Feedback and Hybrid Optimization, coordinating classical and quantum parameter updates to minimize classification errors [8]. 5. Classification Decision and Feature Visualization, where final results are decoded back to the classical domain, demonstrating strong inter-class separability [9]. Future Directions - WiMi plans to expand the LCQHNN model into multimodal learning, integrate with quantum support vector machines and quantum convolutional networks, and promote prototype deployment on quantum hardware to validate performance in real-world scenarios [10]. - The company aims to combine quantum parallel optimization with federated learning frameworks to develop secure and efficient quantum intelligent systems [10]. Company Background - WiMi Hologram Cloud Inc. specializes in holographic cloud services, focusing on various professional applications including AR technologies, holographic devices, and metaverse solutions [12].

Core Viewpoint - WiMi Hologram Cloud Inc. has launched a new Quantum Convolutional Neural Network for Multi-Channel Supervised Learning (MC-QCNN), which allows for efficient processing of multi-channel data and offers significant advantages in various industries such as image classification, medical imaging, and video analysis [1][5]. Group 1: Technological Breakthrough - The MC-QCNN technology features a fully hardware-adaptable quantum convolution kernel design, which enables the processing of multi-channel data efficiently [1][2]. - The architecture includes a systematic design scheme that incorporates convolution kernel structure, qubit layout, and channel interaction encoding, allowing for robust performance against quantum decoherence [2][3]. - WiMi's quantum convolution kernels utilize single-bit rotation gates and controlled parameterized gates, enabling the model to learn complex multi-channel correlations through quantum superposition and entanglement [2][3]. Group 2: Training and Performance - The training framework combines classical and quantum computing, where the classical module handles loss function calculations and the quantum module manages forward propagation and state evolution [3][5]. - The model captures nonlinear correlations between multiple channels, enhancing its ability to recognize joint features in data, such as color distribution patterns in RGB images [3][4]. - Experimental results indicate that the new pooling structure is more stable than traditional methods, maintaining a higher feature retention rate [3]. Group 3: Future Developments - WiMi plans to refine its technology by developing more efficient quantum convolution kernel structures and exploring integration with models like Transformer for processing multimodal data [6]. - The company envisions that quantum deep learning will evolve beyond small-scale tasks to become a significant component in next-generation general AI models [6]. - The combination of quantum computing and artificial intelligence is expected to be a core trend in technological development over the next decade [6]. Group 4: Company Overview - WiMi Hologram Cloud Inc. specializes in holographic cloud services, focusing on areas such as in-vehicle AR holographic HUD, 3D holographic pulse LiDAR, and metaverse holographic AR/VR devices [7]. - The company provides a comprehensive range of holographic AR technologies, including software development and interactive virtual communication solutions [7].

Core Insights - WiMi Hologram Cloud Inc. has developed a hybrid quantum-classical deep learning technology called QB-Net, which integrates lightweight quantum computing modules into the classical U-Net architecture, achieving a reduction in parameters by up to 30 times while maintaining comparable performance [1][9][10] Technology Overview - The core advantage of quantum computing is its ability to express high-dimensional information and perform operations in exponentially dimensional spaces, surpassing classical architectures [2] - WiMi's approach focuses on constructing quantum enhancement modules rather than fully quantized AI models, addressing the bottleneck layer of deep networks with quantum states that can express high-dimensional features efficiently [3][4] - QB-Net retains the U-Net structure but replaces traditional convolutional layers at the bottleneck with a quantum feature compression-transformation-reconstruction module, which consists of three key steps: encoding classical features into quantum states, transforming features through quantum circuits, and decoding quantum states back into classical tensors [5][8] Quantum Feature Transformation - The encoding step minimizes qubit usage while preserving key information, using techniques like linear projection or amplitude encoding [5] - The transformation step utilizes parameterized quantum circuits (PQC) to achieve expressive transformations with significantly fewer parameters compared to traditional convolutional layers [6][7] - The decoding step reconstructs the quantum measurement results into classical tensors, allowing for efficient filtering and abstraction of high-dimensional information [8] Strategic Implications - The introduction of QB-Net signifies a significant advancement in quantum AI technology, demonstrating the real value of quantum computing and its potential integration with deep learning [9] - The hybrid quantum-classical architecture is expected to become a mainstream form of AI, providing a new optimization paradigm for the global AI industry and enhancing enterprise-level intelligent systems [10]

Core Insights - WiMi Hologram Cloud (WIMI) has shown a stock price increase of 2.88% on December 16, reaching $2.86 per share, with a total market capitalization of $36.78 million [1] - For the fiscal year ending December 31, 2024, WIMI reported total revenue of 542 million RMB, a year-over-year decrease of 7.42%, while net profit attributable to shareholders was 71.64 million RMB, reflecting a significant year-over-year increase of 117.01% [1] - The company is registered in the Cayman Islands and operates primarily through its domestic subsidiary, Beijing WiMi Hologram Cloud Software Co., Ltd. [1] Company Overview - WiMi focuses on computer vision holographic cloud services and aims to become a leading and internationally influential holographic cloud platform in China [1] - The company provides a comprehensive range of services in holographic AR technology, including holographic computer vision AI synthesis, visual presentation, interactive software development, online and offline advertising, SDK payment solutions, 5G holographic communication software, facial recognition, and AI face-swapping [1] - WiMi has established itself as one of the largest comprehensive technology solution providers in the holographic cloud sector in China [1] Industry Developments - WIMI has made significant breakthroughs in various holographic application fields such as advertising, entertainment, education, and 5G communications [2] - The company is committed to deep research and market application across all stages of holographic 3D computer vision, including collection, AI synthesis, transmission, and presentation [2] - WIMI aims to build a strong, open service platform that bridges the application of holographic technology and computer vision presentation, promoting cross-industry development [2]

Core Viewpoint - WiMi Hologram Cloud Inc. is advancing its research in image classification through the development of a shallow hybrid quantum-classical convolutional neural network (SHQCNN) model, which integrates innovative quantum computing techniques to enhance performance in this field [1][4]. Group 1: SHQCNN Model Development - The SHQCNN model utilizes an enhanced variational quantum method, optimizing traditional approaches to improve efficiency in image classification tasks [2]. - The model employs a kernel encoding method in the input layer, which enhances data distinction by mapping original image data from low-dimensional to high-dimensional feature space [3]. - Variational quantum circuits are designed in the hidden layer to reduce computational complexity while effectively extracting image features [3]. Group 2: Training and Optimization Techniques - The output layer of the SHQCNN model uses a mini-batch gradient descent algorithm, which improves parameter training speed and model adaptability to changes in training data [4]. - The integration of advanced technologies such as enhanced variational quantum methods, kernel encoding, variational quantum circuits, and mini-batch gradient descent contributes to the model's stability, accuracy, and generalization capabilities [4]. Group 3: Future Potential - The continuous development of quantum computing technology and the expansion of application scenarios suggest that the SHQCNN model will have significant potential across various fields beyond image classification [4].

Core Viewpoint - WiMi Hologram Cloud Inc. has developed a disruptive single-qubit quantum neural network technology that integrates quantum computing with artificial intelligence, addressing the limitations of traditional neural networks in multi-class classification tasks [1][4]. Group 1: Technology Development - The single-qubit quantum neural network technology (SQ-QNN) demonstrates the feasibility of high-dimensional quantum systems for efficient learning, providing a pathway for deep integration of quantum computing and AI [1][3]. - This technology addresses the challenges of training large neural networks, which often require billions of parameters and significant data center resources, leading to high power consumption and costs [2][3]. Group 2: Quantum Neural Network Advantages - Quantum Neural Networks (QNN) utilize qubits and quantum multi-level systems to represent high-dimensional data spaces, overcoming the resource limitations of classical computing [3]. - Compared to traditional deep learning, QNN can achieve complex mappings with shallow quantum circuits, enhancing model compactness and computational efficiency [3][4]. Group 3: Implementation Details - The SQ-QNN technology uses a single high-dimensional qudit to handle multi-class classification tasks, simplifying the model structure compared to classical neural networks [5][6]. - Each category corresponds to a dimension of the quantum system, and the classification process is executed through a high-dimensional unitary operator, reducing circuit depth and training overhead [6][8]. Group 4: Training Methodology - WiMi's technology incorporates a hybrid training method that combines extended activation functions with Support Vector Machine (SVM) optimization, enhancing the network's nonlinear representational capabilities [7][10]. - The training framework allows for efficient parameter optimization and quick convergence to global optimal solutions, effectively sharing the burden on quantum hardware [10][11]. Group 5: Company Overview - WiMi Hologram Cloud Inc. specializes in holographic cloud services, focusing on various applications such as in-vehicle AR holographic HUD, 3D holographic pulse LiDAR, and metaverse holographic technologies [12].

Core Viewpoint - WiMi Hologram Cloud Inc. is actively exploring Quantum Dilated Convolutional Neural Networks (QDCNN) technology, which aims to overcome the limitations of traditional convolutional neural networks (CNNs) in processing complex data and high-dimensional problems, potentially leading to advancements in various fields such as image recognition, data analysis, and intelligent prediction [1][3]. Group 1: Traditional CNN Limitations - Traditional CNNs face bottlenecks in computational efficiency and feature extraction capabilities due to the explosive growth of data volume and increasing problem complexity [2]. - The architecture of traditional CNNs includes convolutional layers, pooling layers, and fully connected layers, which automatically extract features from large datasets [2]. Group 2: Quantum Computing Integration - QDCNN technology integrates quantum computing advantages into traditional CNN architecture, utilizing quantum processors for certain computational operations, which significantly accelerates feature extraction [3][4]. - Quantum computing allows for parallel processing of multiple data states, enhancing the network's ability to capture complex relationships within the data [3][5]. Group 3: Enhanced Feature Extraction - QDCNN not only extracts features obtainable by traditional CNNs but also reveals hidden quantum-level feature information, improving generalization capabilities and reducing overfitting when encountering new data [5][6]. - The use of dilated convolution technology in QDCNN expands the receptive field of the convolution kernel, allowing for better contextual information acquisition without increasing parameters [4]. Group 4: Future Development and Applications - WiMi plans to optimize data transmission and task scheduling between quantum and classical computing to enhance overall operational efficiency [6]. - QDCNN technology is expected to find applications in various fields, including medical research for drug development, intelligent transportation for traffic prediction, and environmental protection for climate change analysis [7][8]. Group 5: Company Overview - WiMi Hologram Cloud Inc. focuses on holographic cloud services, covering areas such as in-vehicle AR holographic HUD, 3D holographic pulse LiDAR, and metaverse holographic AR/VR devices [9].

Financial Performance - Total operating revenues for the six months ended June 30, 2025, were RMB 188,240,855, a decrease of 35.3% compared to RMB 290,815,771 in the same period of 2024[5] - Gross profit for the six months ended June 30, 2025, was RMB 52,095,196, resulting in a gross margin of 27.7%[5] - Net income attributable to WIMI Hologram Cloud, Inc. for the six months ended June 30, 2025, was RMB 106,297,786, compared to RMB 7,663,918 in the same period of 2024, representing a significant increase[5] - The company reported a comprehensive income of RMB 109,147,739 for the six months ended June 30, 2025, compared to a comprehensive loss of RMB 42,429,344 in the same period of 2024[5] - Net income for the six months ended June 30, 2025, was RMB 126,273,050, a significant increase from RMB 12,288,585 in the same period of 2024[9] - Cash provided by operating activities for the six months ended June 30, 2025, was RMB 213,669,364, compared to RMB 150,230,330 in 2024, reflecting a growth of approximately 42%[9] - Basic earnings per share for the six months ended June 30, 2025, were RMB 10.32, compared to RMB 0.78 for the same period in 2024[6] Assets and Liabilities - Total current assets as of June 30, 2025, were RMB 3,327,364,606, an increase of 65.3% from RMB 2,013,304,853 as of December 31, 2024[2] - Total liabilities decreased to RMB 972,862,068 as of June 30, 2025, from RMB 767,423,442 as of December 31, 2024, indicating improved financial stability[2] - The company’s cash and cash equivalents increased to RMB 1,959,991,776 as of June 30, 2025, up from RMB 1,070,513,011 as of December 31, 2024, reflecting a strong liquidity position[2] - The company’s total shareholders' equity increased to RMB 2,425,749,494 by June 30, 2025, compared to RMB 872,469,629 in June 30, 2024, showing robust growth[9] - As of June 30, 2025, total property, plant, and equipment, net amounted to RMB 58,339,212 (USD 8,149,529), down from RMB 125,811,741 (USD 17,646,000) as of December 31, 2024, reflecting a significant decrease[135] Investments and Capital - Short-term investments increased to RMB 1,229,636,750 as of June 30, 2025, up from RMB 847,927,125 at the end of 2024[132] - Gain from sales of investments amounted to approximately RMB 29,784,104 (USD 4,160,605) for the six months ended June 30, 2025[133] - Capital contribution from noncontrolling interests increased to RMB 541,913,983 in 2024, indicating strong investor confidence[9] - The company’s total additional paid-in capital rose to RMB 2,226,171,080 by June 30, 2025, compared to RMB 1,651,937,055 in June 30, 2024[9] Research and Development - Research and development expenses for the six months ended June 30, 2025, were RMB 33,470,081, a decrease of 55.8% compared to RMB 75,820,156 in the same period of 2024[5] - The company’s stock compensation expenses for the six months ended June 30, 2025, amounted to RMB 22,189,939, reflecting ongoing investment in employee incentives[9] Revenue Recognition - The Company recognizes revenue from AR advertising display services at a point in time when the related services have been delivered, based on specific contract terms[80] - The Company follows a five-step model for revenue recognition, which did not result in significant changes in the way revenue is recorded compared to previous guidance[79] - The Company recognizes revenue when an end user completes a transaction, with service fees generally billed monthly on a per transaction basis[85] Joint Ventures and Subsidiaries - On August 21, 2020, Wimi HK established a joint venture, VIDA Semicon Co., Limited, with a 53% equity interest to develop holographic AR technologies in the semiconductor industry[17] - On April 15, 2021, Wimi HK set up another joint venture, Viru Technology Limited, holding a 55% equity interest to focus on AR services[18] - VIYI Technology Inc. was established on September 24, 2020, to accelerate AI algorithm and cloud computing services development[20] - VIYI acquired 100% equity of Fe-da Electronics on September 28, 2020, to enhance its computer chip and intelligent chip business[21] Risks and Concentration - One customer accounted for 13.50% of total revenues for the six months ended June 30, 2025, while another accounted for 11.23%, indicating increased customer concentration risk[155] - For the six months ended June 30, 2025, three vendors accounted for 16.62%, 12.67%, and 11.45% of total purchases, highlighting vendor concentration risk[156] Tax and Compliance - The company reported a current income tax credit of RMB (2,290,252) (USD 319,930) for the six months ended June 30, 2025, compared to RMB (457,941) for the same period in 2024[146] - The company has a deferred tax asset of RMB 440,346 (USD 61,513) as of June 30, 2025, which reflects the allowance for expected credit loss[147] Other Financial Metrics - The company reported a foreign currency translation adjustment loss of RMB 17,125,311 for the six months ended June 30, 2025[5] - The company experienced a foreign currency translation loss of RMB 17,125,311 for the six months ended June 30, 2025[10] - The average translation rates applied to statement of income accounts for the six months ended June 30, 2025, were RMB 1.00 to USD 0.1392[64]

Group 1: Microsoft AI Developments - Microsoft has launched two self-developed AI models: MAI-Voice-1 and MAI-1-preview, marking a significant breakthrough in its AI research [1] - The MAI-Voice-1 model can generate up to one minute of audio content using a single GPU, showcasing its potential in various applications such as real-time news reporting and podcast-style conversations [1] - The MAI-1-preview model is currently in public testing on the LMArena platform and aims to enhance the capabilities of the Copilot assistant, reducing reliance on OpenAI's large language models [1] Group 2: Google DeepMind Innovations - Google DeepMind has introduced the Gemini 2.5 Flash image editing model, which can accurately modify images based on text instructions while maintaining consistency in the appearance of characters and animals [2] - Gemini 2.5 Flash has shown significant improvements in image modification accuracy compared to previous tools and even outperforms the GPT-4 model in several tasks [2][4] Group 3: Apple's AI Acquisition Interests - Apple executives are reportedly in discussions to acquire Mistral, the largest AI startup in Europe, which has raised substantial funding through multiple financing rounds [4] - A successful acquisition would significantly enhance Apple's capabilities and innovation in the AI sector [4] Group 4: WIMI's AI Innovations - WIMI has established a competitive edge in the AI field through an integrated "hardware + software + platform" approach, accelerating the implementation of AI algorithms [6] - The company focuses on combining multimodal large models with spatial computing technology, enabling the native integration of text, images, audio, and video [6] - WIMI is building an open-source ecosystem by providing model codes, computing interfaces, and technical toolchains, facilitating secondary development and commercial validation of vertical models [6]