WiMi Unveils Hybrid Quantum Neural Network for Enhanced Image Classification - Bisinfotech

WiMi Hologram Cloud Inc., a leading global provider of Hologram Augmented Reality (AR) technology, has introduced a hybrid quantum neural network structure (H-QNN) for image multi-classification. This innovative technology seamlessly integrates the spatial feature extraction capabilities of classical convolutional neural networks (CNN) with the high-dimensional nonlinear mapping properties of quantum neural networks (QNN). The result is a hybrid structure that offers enhanced generalization ability and computational efficiency, particularly in multi-class classification scenarios. This development not only optimizes the quantum-classical hybrid learning system but also demonstrates superior classification accuracy and stability compared to similar algorithms in experimental settings, laying a strong foundation for quantum intelligent vision systems. The H-QNN is designed based on the principle that classical networks handle abstraction while quantum networks perform discrimination. The system is structured into three main modules: the feature dimensionality reduction and encoding module, the quantum state transformation module, and the hybrid decision and transfer learning module. First, the feature dimensionality reduction and encoding module leverages the classical convolutional neural network (CNN) structure to extract low-dimensional feature representations of images through multiple convolutional and pooling layers. The feature vectors are then subjected to PCA dimensionality reduction, standardized, and input into the quantum encoding circuit. During this stage, WiMi employs an improved angle encoding method (Angle Embedding) to map real-valued features to quantum state amplitudes. The encoding is made efficient through multi-layer quantum rotation gates (Ry, Rz), reducing quantum gate depth and minimizing encoding noise. Next, the quantum state transformation module performs the essential task of high-dimensional feature mapping and nonlinear discrimination. This module consists of multiple quantum circuit layers, each incorporating parameterized rotation gates and controlled entanglement gates (CNOT or CZ), enabling the quantum states to form nonlinear coupling and entanglement. To mitigate gradient vanishing, WiMi employs a reconfigurable parameter-sharing strategy, allowing different quantum layers to share some trainable parameters. Mixed-state perturbations are introduced to maintain gradient balance during training, effectively preventing the barren plateau phenomenon and ensuring stable convergence in multi-class tasks. The final module, the hybrid decision and transfer learning module, integrates the quantum computing outputs with the classical decision layer. The quantum circuit’s measurement probability distribution is converted into feature vectors and combined with the output from the classical fully connected layer. This fused vector is input into the Softmax layer for the final classification decision. To improve generalization performance in multi-class tasks, WiMi introduces a transfer learning mechanism, enabling the migration of pre-trained quantum layer parameters from small-sample tasks to new tasks. This reduces the number of training epochs and enhances model stability. In practice, this structure can run in both simulation environments and on hardware quantum processing units (QPU). High-performance GPU clusters are used for training the classical modules, while the quantum modules are executed on quantum simulators or FPGA-accelerated quantum kernel estimation environments. This enables a heterogeneous collaboration between classical and quantum computing resources. The key innovations in this technology are reflected in the following aspects: Architectural Design: The H-QNN integrates convolutional neural networks (CNN) and quantum neural networks (QNN) in a deep and efficient manner. Unlike traditional quantum hybrid models, which simply add quantum parts as a classification head, this hybrid model employs a three-stage distributed structure of “convolutional feature extraction—quantum mapping—hybrid decision-making,” enabling quantum networks to not only handle nonlinear discrimination but also reconstruct information at the feature space level. Encoding Strategy: WiMi’s combined angle encoding and principal component analysis (PCA) approach addresses the quantum encoding dimension limitations. By optimizing the cumulative variance contribution rate of PCA, it ensures high fidelity in mapping input features to quantum state amplitudes, maximizing the quantum information utilization. Training Strategy: WiMi incorporates a transfer learning mechanism and parameter-sharing structure, solving issues of gradient vanishing and overfitting that often plague traditional quantum neural networks in multi-class classification tasks. Parameter sharing ensures balanced gradient flow, while transfer learning accelerates convergence on new tasks with fewer epochs. Additionally, WiMi uses an early stopping strategy based on the quantum fidelity metric, monitoring quantum state evolution stability to avoid overfitting. System Implementation: The system employs a heterogeneous computing architecture, running the classical computing part on CPU/GPU platforms while executing the quantum part in quantum simulation modules on FPGA. The FPGA module provides reconfigurable execution logic for parameterized quantum circuits, achieving quantum state updates within nanosecond-level response times, thus significantly enhancing the overall training speed compared to pure CPU or GPU simulations. WiMi’s hybrid quantum neural network structure represents a pivotal step in the transition of quantum artificial intelligence research from theoretical exploration to practical application. The technology showcases the potential of quantum computing in machine learning and offers a strategic solution to the performance limitations of current quantum hardware. By embedding trainable quantum layers within classical neural networks, WiMi effectively harnesses quantum computing resources, enabling real-world visual tasks. This marks the beginning of quantum intelligence moving from the lab to real-world applications, where quantum technology will play a key role in industrial upgrades and expanding human cognition, integrating seamlessly with fields like deep learning, computer vision, and edge computing. Follow the Bisinfotech WhatsApp channel to stay updated on India's new technology ecosystem TagsDeep learning FPGA Quantum Computing quantum networks Quantum Neural Network WiMi