HLHung-Wei Li

Hung-Wei Li

*This is an algorithmically generated author record based on publicly available information.

Current Organizations / Institutes

The organizations/institutes from the scholar's latest article


Hung-Wei Li has a total of 29 co-authors. Here are the top co-authors by number of shared publications:


Latest Publications

  • Wavelet-Guided Context Modeling and Aggregation Network for Anti-UAV Detection

    by Jiaqi Liu, Shunxiang Zhang, Zhengyan Sun, Jixu Zhang, Siyang Zhang, Meng Chen, Hung-Wei Li, Kuan-Ching Li

  • Edge-Based Attitude Estimation for AUVs in Resource-Constrained IoUT Networks: A Kernelized IMSB Approach

    by Xiaojun Mei, Xuran Cao, Huafeng Wu, Jiangfeng Xian, Dezhi Han, Hung-Wei Li, Kuan-Ching Li

    The evolution of the Internet of Underwater Things has positioned Autonomous Underwater Vehicles as critical mobile edge nodes, yet the adverse underwater acoustic communication environment, characterized by high latency and low bandwidth, severely constrains the performance of collaborative sensing. To address these challenges, we propose the Kernelized Intrinsic McAulay-Seidman Bound as an online proxy for network Quality of Service. This metric overcomes the optimal test point selection difficulty and the O(L3) computational complexity inherent in the standard Intrinsic McAulay-Seidman Bound. The K-IMSB framework reformulates the discrete problem into a continuous functional optimization within a Reproducing Kernel Hilbert Space. By leveraging variational methods and a heat kernel on SO(3), we derive a closed-form expression that ultimately involves solving an ill-posed Fredholm integral equation of the first kind. To efficiently solve this, a Recursive Ridge Leverage Score Nystr¨om approximation algorithm is introduced, enabling lightweight, energy-efficient computation on the edge. This algorithm utilizes statistical leverage scores to adaptively identify critical manifold regions, thereby solving the dual challenges of operator discretization and numerical instability. Simulation results for an AUV attitude estimation scenario with Out-of-Sequence Measurements demonstrate that the K-IMSB provides a tight lower bound, and the RLS-Nyström method improves computational efficiency by approximately 21.17%, achieving real-time feasibility for IoUT edge deployment.


  • Multivariate Fractal Autoencoder (MFAE): Sparse Sensor Placement via Cross-Variable Synergy for Ocean Data Reconstruction

    by Qiannan Zhang, Huafeng Wu, Jiangfeng Xian, Xiaojun Mei, Yuxuan Zhang, Li’NIan Liang, Hung-Wei Li, Kuan-Ching Li

    Optimizing sensor placement is crucial for enhancing the coverage and data-acquisition efficiency of ocean monitoring systems. Traditional approaches primarily rely on univariate ocean data for sensor placement, failing to capture the multidimensional coupling characteristics of the ocean environment, while the potential of multivariate datasets remains underexplored. To address this limitation, this work proposes an innovative multivariate fractal autoencoder (MFAE) framework that leverages multivariate data to solve the sparse sensor placement problem. The MFAE optimizes sensor placement by dynamically updating multivariate feature weights and extracting latent spatial correlations. Furthermore, by optimizing feature weight initialization and enhancing autoencoder training protocols, we propose an entropy-weighted MFAE (EnMFAE) to establish an accurate nonlinear mapping between low-dimensional sampling spaces and full-state reconstructions. Validation experiments are conducted on temperature and salinity datasets from the North Pacific and Arctic Oceans, and the results demonstrate the superior performance of MFAE and EnMFAE relative to the proper orthogonal decomposition (POD), QR, and random placement baselines. With only ten selected sensors, the MFAE achieves average reconstruction error reductions of 2.96% (for temperature) and 2.12% (for salinity) in the North Pacific, and 5.78% (for temperature) and 7.71% (for salinity) in the Arctic, respectively, significantly outperforming the compared random placement method with decoder-based reconstruction. MFAE offers a novel paradigm for optimizing sensor networks in complex ocean environments by leveraging multivariate data reconstruction.

  • An asymmetric graph neural network with time-position encoding for link prediction in dynamic networks

    by Lihong Han, Mingming Li, Meng-Yen Hsieh, Hung-Wei Li, Rubén González Crespo, Kuan-Ching Li

    In dynamic networks, the network structure continuously changes over time, such as in social networks and transportation systems. Accurately predicting links in dynamic networks is an extremely difficult task. Although existing Temporal Graph Neural Networks (TGNNs) have demonstrated considerable success in predicting links in dynamic networks, their symmetric evaluation strategy limits their scalability in dynamic scenarios. For example, in full-node evaluation, these symmetric update methods not only require simultaneous updates of both source and target nodes but also trigger cascading updates in the neighborhood context, resulting in significant computational overhead. To control this cost, existing TGNNs models have to compromise between temporal proximity and topological structure: either truncate historical context leading to the loss of long-term dependencies, or simplify neighborhood sampling resulting in over-smoothing. This trade-off severely limits model performance.To address the above challenges, this paper propose the TPGNN model, which is an asymmetric temporal graph neural network that only updates source nodes. During evaluation, by updating only the source nodes and precomputing and freezing the target nodes and candidates, it eliminates cascading updates while also improving performance. Additionally, the model introduces inter-layer smoothing, residual connections, multi-scale spatio-temporal encoding, and position-aware attention modules. In link prediction experiments, TPGNN achieves significant performance improvements on four public datasets, while maintaining CPU-friendly efficiency, enabling real-time evaluation in resource-constrained scenarios. The code has been released at: https://github.com/LiMinlr/TPGNN .


  • Open Access

    Cross-Domain Generalization for LiDAR-Based 3D Object Detection in Infrastructure and Vehicle Environments

    by Peng Zhi, Longhao Jiang, Xiao Yang, Xingzheng Wang, Hung-Wei Li, Qingguo Zhou, Kuan-Ching Li, Mirjana Ivanović

    In the intelligent transportation field, the Internet of Things (IoT) is commonly applied using 3D object detection as a crucial part of Vehicle-to-Everything (V2X) cooperative perception. However, challenges arise from discrepancies in sensor configurations between vehicles and infrastructure, leading to variations in the scale and heterogeneity of point clouds. To address the performance differences caused by the generalization problem of 3D object detection models with heterogeneous LiDAR point clouds, we propose the Dual-Channel Generalization Neural Network (DCGNN), which incorporates a novel data-level downsampling and calibration module along with a cross-perspective Squeeze-and-Excitation attention mechanism for improved feature fusion. Experimental results using the DAIR-V2X dataset indicate that DCGNN outperforms detectors trained on single datasets, demonstrating significant improvements over selected baseline models.