MicroCloud Hologram Inc. Releases Next-Generation Quantum Convolutional Neural Network Multi-Class Classification Technology, Driving Quantum Machine Learning Towards Practicalization

Rhea-AI Summary

MicroCloud Hologram (NASDAQ: HOLO) announced a next-generation quantum convolutional neural network (QCNN) multi-class classification method using a hybrid quantum-classical training framework on Nov 14, 2025.

The system encodes MNIST samples using 8 qubits plus 4 auxiliary qubits, introduces a quantum perceptron and optimized entanglement layers, and reports QCNN accuracy comparable to classical CNNs at the same parameter scale for a four-class task. The company also disclosed cash reserves >3 billion RMB and plans to invest over $400 million in frontier tech including quantum computing.

Positive

• Cash reserves exceeding 3 billion RMB
• Planned investment of over $400 million into frontier technologies
• QCNN achieved comparable accuracy to classical CNNs on four-class MNIST
• Model implemented with 8 qubits + 4 auxiliary qubits

Negative

• Experimental validation limited to four classes from MNIST
• Performance reliant on NISQ-era hardware with noise and scalability limits
• No quantified commercial deployment timeline or revenue impact disclosed

Insights

Quantum ML Analyst

HOLO announces a QCNN hybrid quantum-classical multi-class classifier with MNIST 4-class parity; technical proof-of-concept, limited near-term commercial impact.

HOLO's announcement describes a hybrid quantum-classical training loop that encodes partial MNIST samples into eight qubits plus four auxiliaries, applies a quantum convolution module and a proposed "quantum perceptron," then feeds measurement probabilities through a softmax and cross-entropy loss to a classical optimizer. The described mechanism preserves convolutional logic while replacing classical activation with quantum state evolution and uses circuit-level optimizations to limit gate count and entanglement depth for NISQ-era feasibility. The company reports comparable accuracy to classical CNNs on a four-class MNIST task and positions this as a step toward scaling quantum machine learning.

Dependencies and risks are explicit in the text: model utility hinges on quantum hardware improvements (more qubits, lower noise, higher fidelity) and on scaling the circuit design beyond the 8+4 qubit experimental setup. The release also cites a planned capital allocation from cash reserves ("cash reserves exceeding 3 billion RMB" and a plan to invest "more than 400 million in USD"), which is a stated resource commitment but not a technical guarantee. Watch implementation milestones such as broader benchmark results beyond a four-class MNIST split, demonstrations on larger datasets, and any disclosed timelines for scaling or hardware partnerships; the announcement date is November 14, 2025.

SHENZHEN, China, Nov. 14, 2025 /PRNewswire/ -- MicroCloud Hologram Inc. (NASDAQ: HOLO), ("HOLO" or the "Company"), a technology service provider, has launched a groundbreaking technological achievement—a multi-class classification method based on the Quantum Convolutional Neural Network (QCNN) using hybrid quantum-classical learning. This method not only demonstrates the enormous potential of quantum computing in image recognition and complex classification tasks but also provides a new path for the development of artificial intelligence in the post-Moore era.

The launch background of this technology lies in the rapid popularization of deep learning in fields such as computer vision, speech recognition, and natural language processing, where classical neural networks are gradually encountering bottlenecks in computing power, energy consumption, and model complexity. Especially under the trend of continuously expanding data scales and increasing number of categories in classification tasks, the limitations of traditional computing architectures are becoming increasingly evident. At the same time, the rise of quantum computing provides unprecedented possibilities for breaking this bottleneck. Quantum computers utilize quantum characteristics such as superposition and entanglement to achieve parallel computing in an exponentially large computational space, and their advantages in combinatorial optimization, matrix operations, and probability distribution sampling highly align with the needs of machine learning.

The core of HOLO's this technology is a multi-class classification model that combines quantum convolutional neural networks with a hybrid quantum-classical optimization framework. The research team, based on the TensorFlow Quantum platform, has built a training mechanism that integrates quantum circuits and classical optimizers. In terms of input data, selected partial samples from the MNIST dataset, especially four types of handwritten digit images among them, as training and validation objects. Data encoding is completed through eight qubits, supplemented by four auxiliary qubits to support the computation and optimization process, forming a quantum computing framework that combines efficiency and scalability.

In terms of model design, HOLO proposed a brand-new quantum perceptron model. This model takes quantum state evolution and measurement as its core, introducing the feature extraction concept of convolutional neural networks into the quantum circuit structure. Unlike traditional neurons that rely on nonlinear activation functions to model complex patterns, the quantum perceptron naturally forms high-dimensional feature mappings using the superposition and entanglement effects of quantum gates, possessing the capability to express complex functions within a smaller parameter space. Further circuit optimizations include reducing redundant gate operations, improving the entanglement structure between layers, and introducing parameterized rotation gates after the convolutional layer to enhance nonlinear feature extraction, thereby ensuring that under the hardware-limited conditions of the NISQ (Noise Intermediate-Scale Quantum Computing) era, the model can still maintain good expressiveness and stability.

In the training process, the hybrid quantum-classical learning mechanism plays a key role. The quantum circuit is responsible for quantum state encoding and evolution of input samples, and outputs the measurement results as a quantum probability distribution; these results are then passed to the classical computing unit, normalized through the softmax activation function, and finally form classification probabilities. Subsequently, the system uses the cross-entropy loss function to measure the gap between the prediction results and the true labels, and iteratively updates the quantum circuit parameters through a classical optimizer. This design takes into account the advantages of quantum computing in feature modeling and the mature experience of classical computing in optimization algorithms, thereby significantly improving training efficiency and model convergence speed.

Experimental results indicate that in the task scenario of four-class classification, the performance of HOLO's quantum convolutional neural network is comparable in accuracy to that of classical convolutional neural networks under the same parameter scale. This conclusion not only proves the feasibility of quantum neural networks in practical tasks but also further reinforces the value of quantum machine learning as a future technological direction.

In terms of the technical implementation logic, this achievement mainly consists of three core stages: first is data encoding, where HOLO uses amplitude encoding to map MNIST images onto eight qubits, while utilizing auxiliary qubits to handle specific feature extraction tasks. Second is the quantum convolution module, which achieves the extraction of local features and the combination of global features through the arrangement of quantum gates and entanglement; this process is similar to the convolution kernel and pooling operations in classical convolutional networks, but manifests as higher-dimensional state evolution in the quantum state space. Finally, in the classification output stage, the probability distribution obtained from quantum measurements enters the softmax layer, and the rotation parameters of the quantum gates are continuously adjusted through the hybrid optimization framework, thereby gradually approaching the optimal solution. The overall process not only retains the logical structure of convolutional neural networks but also fully leverages the parallel computing advantages of quantum superposition states.

HOLO's this technology is not merely a simple model migration but an innovative achievement after deep optimization at the quantum circuit level. The proposal of the quantum perceptron effectively controls the circuit complexity, avoiding the noise accumulation issues caused by redundant gate operations; the optimized entanglement layer structure significantly enhances the model's expressive power, enabling it to capture more complex correlations between data. These innovation points lay a solid foundation for quantum neural networks in future large-scale practical applications.

From an industry background perspective, multi-class classification tasks are widely present in scenarios such as computer vision, medical image analysis, speech recognition, natural language processing, financial risk control, and more. Traditional deep learning methods have achieved tremendous accomplishments in these fields, but their high energy consumption, long training times, and strong dependence on computing resources are gradually becoming constraining factors. The quantum convolutional neural network method launched by HOLO was born precisely in the context of addressing these challenges. By transplanting classical convolutional structures into a quantum computing framework, this method not only reduces the computational complexity during model training but also provides the possibility for achieving true breakthroughs in computing power in the future as quantum hardware conditions gradually mature.

The significance of this technology is not limited to experimental verification on the MNIST dataset but lays the foundation for the application of quantum machine learning in more complex and broader tasks. With the continuous advancement of quantum hardware, more qubits, lower noise levels, and quantum chips with higher fidelity will gradually emerge, and models based on quantum convolutional neural networks are expected to expand to cutting-edge scenarios such as large-scale image recognition, real-time video processing, and natural language multi-class understanding. HOLO also plans to further optimize the scalability of quantum circuits in subsequent research and development, exploring the combination of multi-layer quantum convolutional networks with deep residual structures.

HOLO's quantum convolutional neural network multi-class classification technology based on hybrid quantum-classical learning not only demonstrates the unique advantages of quantum computing in artificial intelligence but also provides new solutions for addressing the bottleneck issues in the development process of deep learning. With the joint progress of future quantum hardware and algorithms, this technology is expected to truly move out of the laboratory and toward industrial applications, becoming an important force in leading the intelligent society.

About MicroCloud Hologram Inc.

MicroCloud Hologram Inc. (NASDAQ: HOLO) is committed to the research and development and application of holographic technology. Its holographic technology services include holographic light detection and ranging (LiDAR) solutions based on holographic technology, holographic LiDAR point cloud algorithm architecture design, technical holographic imaging solutions, holographic LiDAR sensor chip design, and holographic vehicle intelligent vision technology, providing services to customers offering holographic advanced driving assistance systems (ADAS). MicroCloud Hologram Inc. provides holographic technology services to global customers. MicroCloud Hologram Inc. also provides holographic digital twin technology services and owns proprietary holographic digital twin technology resource libraries. Its holographic digital twin technology resource library utilizes a combination of holographic digital twin software, digital content, space data-driven data science, holographic digital cloud algorithms, and holographic 3D capture technology to capture shapes and objects in 3D holographic form. MicroCloud Hologram Inc. focuses on developments such as quantum computing and quantum holography, with cash reserves exceeding 3 billion RMB, and plans to invest more than 400 million in USD from the cash reserves to engage in blockchain development, quantum computing technology development, quantum holography technology development, and derivatives and technology development in frontier technology fields such as artificial intelligence AR. MicroCloud Hologram Inc.'s goal is to become a global leading quantum holography and quantum computing technology company. For more information, please visit http://ir.mcholo.com/

Safe Harbor Statement

This press release contains forward-looking statements as defined by the Private Securities Litigation Reform Act of 1995. Forward-looking statements include statements concerning plans, objectives, goals, strategies, future events or performance, and underlying assumptions and other statements that are other than statements of historical facts. When the Company uses words such as "may," "will," "intend," "should," "believe," "expect," "anticipate," "project," "estimate," or similar expressions that do not relate solely to historical matters, it is making forward-looking statements. Forward-looking statements are not guarantees of future performance and involve risks and uncertainties that may cause the actual results to differ materially from the Company's expectations discussed in the forward-looking statements. These statements are subject to uncertainties and risks including, but not limited to, the following: the Company's goals and strategies; the Company's future business development; product and service demand and acceptance; changes in technology; economic conditions; reputation and brand; the impact of competition and pricing; government regulations; fluctuations in general economic; financial condition and results of operations; the expected growth of the holographic industry and business conditions in China and the international markets the Company plans to serve and assumptions underlying or related to any of the foregoing and other risks contained in reports filed by the Company with the Securities and Exchange Commission ("SEC"), including the Company's most recently filed Annual Report on Form 10-K and current report on Form 6-K and its subsequent filings. For these reasons, among others, investors are cautioned not to place undue reliance upon any forward-looking statements in this press release. Additional factors are discussed in the Company's filings with the SEC, which are available for review at www.sec.gov. The Company undertakes no obligation to publicly revise these forward-looking statements to reflect events or circumstances that arise after the date hereof.

 

View original content:https://www.prnewswire.com/news-releases/microcloud-hologram-inc-releases-next-generation-quantum-convolutional-neural-network-multi-class-classification-technology-driving-quantum-machine-learning-towards-practicalization-302615815.html

SOURCE MicroCloud Hologram Inc.

FAQ

What did MicroCloud Hologram (HOLO) announce on November 14, 2025?

HOLO launched a next-generation QCNN multi-class classification method using hybrid quantum-classical training and an 8-qubit + 4-auxiliary architecture.

How did HOLO validate the QCNN model and what were the results?

HOLO trained and validated the model on selected four classes from MNIST, reporting accuracy comparable to classical CNNs at the same parameter scale.

What quantum resources does HOLO's QCNN use for training?

The implementation uses amplitude encoding with 8 qubits plus 4 auxiliary qubits, optimized entanglement, and parameterized rotation gates.

Does HOLO report financial capacity to develop quantum technologies?

Yes; HOLO disclosed cash reserves >3 billion RMB and plans to invest over $400 million in frontier technology projects.

What are the main limitations of HOLO's announced QCNN technology for investors?

Results are limited to a small-scale MNIST task and remain dependent on NISQ-era hardware improvements for scalability and practical deployment.