WiMi Hologram Cloud Inc. (NASDAQ: WiMi) is a global hologram augmented reality technology provider with a novel deep convolutional neural network designed for image recognition. The company has developed a model that utilizes quantum parameterized circuits as the core computing structure. This is a departure from traditional deep learning approaches and a step toward harnessing the parallel processing capabilities of quantum computing. This technology addresses the limitations of computational complexity, memory consumption, and training efficiency by combining quantum circuits and quantum-classical hybrid training mechanisms. This model uses a quantum parameterization circuit as the core computing structure, performs feature extraction of image data through a quantum convolution layer, and utilizes a quantum classification layer to complete the final recognition task. The quantum feature fusion module can reduce some of the computational overhead compared to traditional methods.
Quantum parameterization circuit for image recognition
WiMi Hologram Cloud Inc. demonstrated advances in quantum deep convolutional neural networks, hinting at the deepening of quantum machine learning applications in artificial intelligence tasks such as image recognition, and also provided new research directions for realizing future large-scale quantum intelligent computing systems. WiMi’s approach centers on a fundamentally different computational structure, a quantum parameterization circuit, as the core of a deep convolutional neural network. The new model utilizes these circuits to perform feature extraction of image data through a quantum convolution layer and finally a quantum classification layer for final recognition. This architecture mirrors the hierarchical structure of classical deep convolutional neural networks, but importantly leverages the parallel computing capabilities of quantum circuits to increase processing speed when processing high-dimensional data.
The system starts by mapping classical image data to quantum state space via a data encoding module. This is a necessary step when quantum computers manipulate quantum state information. The conversion of this pixel information into qubit probability amplitudes is accomplished by methods such as amplitude encoding and angular encoding to prepare the data for quantum circuit processing. Following encoding, a quantum convolution layer extracts features using parameterized quantum gates. This works similarly to the classic network convolution kernel, but may yield higher computational efficiency. At the heart of this layer is a specially designed circuit by WiMi, consisting of rotation gates, control gates, and entanglement gates.
These gates manipulate the evolution of quantum states through trainable parameters. Rotation gates adjust the state angles of the qubits, control gates establish correlations, and entanglement gates create complex entangled structures. The company explains, “These operations allow quantum circuits to form feature extraction capabilities similar to classical convolutional layers, while having higher expressive power.” As these layers are built up, the network progressively extracts higher-level image features, starting with edges and textures and ending with complex shape and structural information. The parallelism of quantum states within the computational process allows the state space to grow exponentially, greatly increasing efficiency. To overcome the limitations of current quantum hardware, WiMi implemented a hybrid quantum-classical training mechanism. This approach utilizes quantum circuits for forward calculations while relying on classical computers to update parameters.
It utilizes the principles of variational quantum algorithms and combines parameterized quantum circuits with classical optimizers to solve complex problems with limited quantum resources. The quantum feature fusion module integrates feature information from different qubits through additional quantum gate operations and completes the information integration through a quantum state evolution process. This reduces some computational overhead.
Quantum convolutional layer architecture and feature extraction
The pursuit of quantum machine learning continues to gain momentum, extending beyond theoretical exploration into the realm of applied technology. WiMi Hologram Cloud Inc. is a global hologram augmented reality technology provider. This development is notable because it comes from a company focused on holographic augmented reality, and it shows that interest in quantum computing is spreading beyond traditional technology and artificial intelligence companies. While many organizations are researching quantum algorithms, WiMi’s approach focuses on implementing concrete architectures, leveraging quantum parameterized circuits as the underlying computing structure for deep learning models. At the heart of this architecture is a layered system that mirrors classical convolutional neural networks but is adapted for quantum processing. This transformation is achieved through techniques such as amplitude encoding and angular encoding, allowing quantum circuits to effectively process image data.
The main advantage lies in the ability of quantum gates to operate on multiple superposition states simultaneously for highly parallel feature extraction when analyzing complex image structures. This parallelism within an exponentially large state space is expected to significantly improve computational efficiency when processing high-dimensional data. This approach leverages the strengths of both computing paradigms. Quantum circuits handle forward calculations, and classical computers manage parameter updates. During training, image data is encoded into quantum states, processed through the circuit, and then analyzed to calculate the error between the network’s output and the true label. A classical optimization algorithm then computes the gradient information and updates the trainable parameters in the quantum circuit to form a collaborative training process. The company says its quantum feature fusion module can reduce some computational overhead compared to traditional methods, highlighting the practical considerations driving the design choice. This model has the potential to provide exponential acceleration in certain tasks compared to traditional deep convolutional neural networks, and the theoretical potential for computational advantages is very high.
WiMi’s technology demonstrates the potential value of quantum computing in the field of artificial intelligence. By combining the physics of quantum computing with the model structure of deep learning, we can build new intelligent systems with higher expressive power and more powerful computational efficiency.
Data encoding and quantum feature fusion process
WiMi Hologram Cloud Inc. is integrating quantum computing into its deep convolutional neural networks. This effort differs from many other companies that focus solely on traditional machine learning approaches. The company, a global hologram augmented reality technology provider, has made progress in developing quantum systems for image recognition. A key element of this progress lies in how classical image data is transformed into quantum form. Following data encoding, WiMi’s quantum convolutional layer performs feature extraction to reflect the functionality of a classical convolutional neural network’s kernel. However, instead of traditional mathematical operations, this layer utilizes a set of parameterized quantum gates that act on local qubits.
WiMi’s research shows that superposition of quantum states can perform highly parallel feature extraction, significantly increasing efficiency when processing complex image structures. “These quantum gates form a function similar to a convolution filter and can perform feature mapping on input quantum states,” the company explains. A key innovation within the network architecture is a quantum functional fusion module designed to integrate information from different qubits. This module employs a quantum entanglement mechanism to fuse image features from different regions to create a high-dimensional representation with enhanced discriminatory power. Unlike traditional neural networks that rely on matrix multiplication for feature fusion, WiMi’s approach completes information integration through quantum state evolution, which can reduce computational overhead to some extent. A subsequent quantum classification layer outputs classification results by measuring the probability distribution of the quantum states. This process is similar to fully connected layers in classical networks, but it is performed in quantum state space.
Hybrid quantum-classical training mechanism for WiMi models
WiMi Hologram Cloud Inc. Central to this progress is a hybrid training mechanism, a practical approach designed to bridge the gap between the limitations of current quantum hardware and the demands of complex machine learning tasks. This transformation allows quantum circuits to effectively process image data. At the heart of a quantum convolutional layer is a specially designed circuit consisting of basic quantum logic gates, rotations, controls, and entanglements that manipulate quantum states through trainable parameters. Realizing this potential will require addressing the limitations of current quantum hardware.
