Insider Brief
- WiMi Hologram is researching the use of neural network models to optimize parameter selection in twin-field quantum key distribution (TF-QKD) systems, aiming to reduce computation time and improve system performance.
- The company evaluated three machine learning models—BPNN, RBFNN, and GRNN—and found that RBFNN and GRNN delivered higher prediction accuracy in high-dimensional parameter spaces, while BPNN offered the fastest computation.
- WiMi said future work will explore advanced AI approaches such as deep learning and reinforcement learning while integrating the technology with quantum communication hardware.
- Image from Unsplash by Alina Grubnyak.
Press release – WiMi Hologram Cloud Inc. (NASDAQ: WiMi) (“WiMi” or the “Company”), a leading global Hologram Augmented Reality (“AR”) Technology provider, announced that they are researching the use of neural networks for machine learning to optimize parameters in the dual-field quantum key distribution (TF-QKD) system. This innovative approach aims to leverage the powerful fitting ability and generalization performance of neural networks to directly predict the optimal parameter configuration for the TF-QKD system, significantly reducing computation time and resource consumption.
In the study, WiMi trained and evaluated three different types of neural network models:
Backpropagation Neural Network (BPNN): Based on the error backpropagation algorithm, BPNN minimizes prediction errors by continuously adjusting the network weights and biases. Due to its flexibility and wide applicability, BPNN has become the preferred model in many fields.
Radial Basis Function Neural Network (RBFNN): Using radial basis functions as activation functions for the hidden layer neurons, RBFNN efficiently handles nonlinear problems and is particularly suitable for high-dimensional data and scenarios requiring high precision.
Generalized Regression Neural Network (GRNN): Based on probability density estimation, GRNN uses kernel function methods to achieve nonlinear regression, excelling in handling small sample data and uncertainty issues.
Through training and testing these three neural network models, WiMi found that all models could accurately predict the optimal parameters of the TF-QKD system to some extent. Among them, RBFNN and GRNN performed especially well in high-dimensional parameter spaces, showing higher prediction accuracy. Compared to LSA, the neural network-based prediction method achieved a significant reduction in computation time, cutting it by multiple orders of magnitude. BPNN, due to its relatively simple structure, had the fastest computation speed; whereas RBFNN and GRNN, though slightly more complex in terms of computational cost, still remained within acceptable limits, and their enhanced prediction accuracy often brought more practical application value.
Considering the varying optimization needs of different TF-QKD systems (such as real-time requirements and precision demands), WiMi also conducted a comprehensive comparison of prediction accuracy and time consumption. The results indicate that for scenarios requiring rapid response with lower precision demands, BPNN is the ideal choice. On the other hand, for applications that prioritize high accuracy and can tolerate certain computation time, RBFNN or GRNN is more suitable.
The main technical advantage of using neural networks for TF-QKD system parameter optimization lies in significantly reducing the computational complexity of parameter optimization, accelerating the key generation rate, and enhancing the system’s real-time responsiveness. Neural networks can automatically learn and adapt to changes in the quantum communication environment, providing the possibility for dynamic adjustment of system parameters. As quantum communication technology develops, neural network models can be further upgraded and optimized to accommodate more complex quantum key distribution protocols and higher security requirements.
In the future, WiMi will continue to deepen its research into neural networks for TF-QKD parameter optimization, exploring more advanced neural network architectures and training strategies, such as deep learning, reinforcement learning, etc., with the aim of achieving more efficient and intelligent quantum key distribution systems. At the same time, it will strengthen integration with quantum communication hardware platforms to promote the practical application and commercialization of quantum communication technologies, contributing to the development of secure and efficient quantum communication networks.
