Hong Kong University of Science and Technology
Hong Kong University of Science and Technology Institutional RepositoryNot a member yet
162821 research outputs found
Sort by
Accelerated China's cold regions contraction under climate warming
No full textCold regions are vital to the Earth system, influencing water storage, energy balance, and ecological stability. China's diverse terrain includes extensive cold regions that are shrinking due to global warming, with profound implications for climate resilience. Despite their importance, comprehensive assessments of these regions’ past trends and future projections are lacking. This study employs an array of data, including gridded daily dataset (CN05.1), China Meteorological Forcing Dataset (CMFD), European Centre for Medium-Range Weather Forecasts (ECMWF) Reanalysis v5-Land (ERA5-Land), and original and bias-corrected outputs from Coupled Model Intercomparison Project Phase 6 (CMIP6) global climate models (GCMs), to analyze the historical (1979–2016) and projected (2015–2100) dynamics of China's cold regions under different emission scenarios. Our findings indicate that from 1979 to 2016, China's cold regions covered an average of 3.87 million km2, or 40.4 % of China's land area, with the most substantial areas located in the Tibetan Plateau, Tianshan and Pamirs mountain ranges, and the northeastern region. These regions experienced a significant decline, at a rate of 154,000 km2 per decade. Future projections based on bias-corrected CMIP6 data suggest an accelerating contraction, with estimates ranging from a decrease of 54,000 km2 per decade under the SSP126 scenario to 210,000 km2 per decade under the SSP585 scenario. By the end of the 21st century, the cold regions’ extent could diminish by more than 50 % under the high-emission SSP585 scenario compared to the historical baseline. The analysis shows that nearly 95 % of variations in the extent of China's cold regions can be attributed to shifts in areas with an annual average temperature at or below 5 °C. These results underscore the need for urgent climate adaptation strategies, particularly as China's cold regions continue to shrink in response to climate warming driven by human-induced emissions.</p
Federated Edge Learning: Algorithms, Architectures and Trustworthiness
No full textThis book presents various effective schemes from the perspectives of algorithms, architectures, privacy, and security to enable scalable and trustworthy Federated Edge Learning (FEEL). From the algorithmic perspective, the authors elaborate various federated optimization algorithms, including zeroth-order, first-order, and second-order methods. There is a specific emphasis on presenting provable convergence analysis to illustrate the impact of learning and wireless communication parameters. The convergence rate, computation complexity and communication overhead of the federated zeroth/first/second-order algorithms over wireless networks are elaborated. From the networking architecture perspective, the authors illustrate how the critical challenges of FEEL can be addressed by exploiting different architectures and designing effective communication schemes. Specifically, the communication straggler issue of FEEL can be mitigated by utilizing reconfigurable intelligent surface and unmanned aerial vehicle to reconfigure the propagation environment, while over-the-air computation is utilized to support ultra-fast model aggregation for FEEL by exploiting the waveform superposition property. Additionally, the multi-cell architecture presents a feasible solution for collaborative FEEL training among multiple cells. Finally, the authors discuss the challenges of FEEL from the privacy and security perspective, followed by presenting effective communication schemes that can achieve differentially private model aggregation and Byzantine-resilient model aggregation to achieve trustworthy FEEL. This book is designed for researchers and professionals whose focus is wireless communications. Advanced-level students majoring in computer science and electrical engineering will also find this book useful as a reference.</p
Editorial Introduction: Quantitative History of China: State Capacity, Institutions and Development
No full textThis introductory chapter sets the stage for this volume. It begins with a brief survey of the growth of scholarship on the quantitative history of China. It then reviews the eleven remaining chapters, organizing them into four parts: War, Technology, and the Needham Puzzle; Rulers, Institutions and the Military; State Capacity, Fiscal and Financial Development; Religion and Agricultural Development. The chapter concludes with a call for further development of the quantitative history of China.</p
Cross-View Generalized Diffusion Model for Sparse-View CT Reconstruction
No full textSparse-view computed tomography (CT) reduces radiation exposure by subsampling projection views, but conventional reconstruction methods produce severe streak artifacts with undersampled data. While deep-learning-based methods enable single-step artifact suppression, they often produce over-smoothed results under significant sparsity. Though diffusion models improve reconstruction via iterative refinement and generative priors, they require hundreds of sampling steps and struggle with stability in highly sparse regimes. To tackle these concerns, we present the Cross-view Generalized Diffusion Model (CvG-Diff), which reformulates sparse-view CT reconstruction as a generalized diffusion process. Unlike existing diffusion approaches that rely on stochastic Gaussian degradation, CvG-Diff explicitly models image-domain artifacts caused by angular subsampling as a deterministic degradation operator, leveraging correlations across sparse-view CT at different sample rates. To address the inherent artifact propagation and inefficiency of sequential sampling in generalized diffusion model, we introduce two innovations: Error-Propagating Composite Training (EPCT), which facilitates identifying error-prone regions and suppresses propagated artifacts, and Semantic-Prioritized Dual-Phase Sampling (SPDPS), an adaptive strategy that prioritizes semantic correctness before detail refinement. Together, these innovations enable CvG-Diff to achieve high-quality reconstructions with minimal iterations, achieving 38.34 dB PSNR and 0.9518 SSIM for 18-view CT using only 10 steps on AAPM-LDCT dataset. Extensive experiments demonstrate the superiority of CvG-Diff over state-of-the-art sparse-view CT reconstruction methods. The code is available at https://github.com/xmed-lab/CvG-Diff.</p
Enzyme-mimetic functional synergistic strategy of Mn-based spinel catalysts: regulating oxygen vacancies and radical pathways to promote photocatalytic methane oxidation to C1-C2 oxygenates
No full textThe photocatalytic conversion of methane into valuable liquid oxygenates presents a promising strategy for advancing a sustainable chemical industry. However, achieving both high conversion efficiency and high selectivity using non-precious metal catalysts under ambient conditions remains a significant challenge. Inspired by the functional characteristics of natural enzymatic catalytic systems, this study has constructed a multicomponent synergistic catalytic system with ZnMn₂O₄ as the support. Through the defective engineering design of fluorine doping and in situ growth of tungsten species, it achieves efficient construction of the microenvironment of active centers and precise regulation of free radical intermediates. The optimized 3W20F-ZnMn₂O₄ catalyst achieved a liquid-phase product yield (including methanol and ethanol, etc.) of 442.02 μmol·g⁻¹ ·h⁻¹ with a selectivity of 98.6 % under ambient temperature and pressure—surpassing the performance of most reported non-precious metal-based catalytic systems. Studies have demonstrated that catalysts doped with the two elements exhibit significant differences in the oxygen sources of products, which reflects the distinct pathways they follow in catalyzing methane oxidation. Specifically, the oxygen vacancy (FVo) sites formed by F doping primarily adhere to a single reaction pathway using water as the oxygen source, where the active oxygen species are mainly dominated by ·OH generated from the dissociation of water molecules, ultimately yielding ethanol and small amount of methanol. In contrast, the active sites formed by W doping involve multiple reaction pathways. The production of ethanol relies more on ·OH from water molecule dissociation, while the generation of methanol and methyl hydroperoxide is more dependent on active oxygen species derived from oxygen activation. This study deepens the understanding of the multi-element synergistic catalytic mechanism and lays a foundation for the development of sustainable and efficient catalytic systems.</p
VAMPIRE: Uncovering Vessel Directional and Morphological Information from OCTA Images for Cardiovascular Disease Risk Factor Prediction
No full textCardiovascular disease (CVD) remains the leading cause of death worldwide, requiring urgent development of effective risk assessment methods for timely intervention. While current research has introduced non-invasive and efficient approaches to predict CVD risk from retinal imaging with deep learning models, the commonly used fundus photographs and Optical Coherence Tomography (OCT) fail to capture detailed vascular features critical for CVD assessment compared with OCT angiography (OCTA) images. Moreover, existing methods typically classify CVD risk only as high or low, without providing a deeper analysis on CVD-related blood factor conditions, thus limiting prediction accuracy and clinical utility. As a result, we propose a novel multi-purpose paradigm of CVD risk assessment that jointly performs CVD risk and CVD-related condition prediction, aligning with clinical experiences. Based on this core idea, we introduce OCTA-CVD, the first OCTA dataset for CVD risk assessment, and a Vessel-Aware Mamba-based Prediction model with Informative Enhancement (VAMPIRE) based on OCTA enface images. Our proposed model aims to extract crucial vascular characteristics through two key components: (1) a Mamba-Based Directional (MBD) Module that captures fine-grained vascular trajectory features and (2) an Information-Enhanced Morphological (IEM) Module that incorporates comprehensive vessel morphology knowledge. Experimental results demonstrate that our method can surpass standard classification backbones, OCTA-based detection methods, and ophthalmologic foundation models. Our codes and the collected OCTA-CVD dataset are available at https://github.com/xmed-lab/VAMPIRE.</p
Automating digitalization of engineered slopes to support interactive digital twin visualization
No full textSlope safety management is transitioning toward the digital twin paradigm, where digital models serve as a base plate for the development of further applications. This paper presents an automated solution to creating digital models for engineered slope digital twins. Utilizing city-scale point cloud, the proposed approach creates lightweight BIM models of engineered slopes consisting of drainage channels and slope topography. The proposed approach leverages a convolutional neural network (CNN) model, DrainNet, for fine-scale drainage channel delineation in images, an automated parametric modeling pipeline for scalable BIM reconstruction, and a web-based digital twin for interactive visualization. Experiments demonstrate that DrainNet achieves state-of-the-art 81.26 % Dice with only 0.28 M parameters on a domain-knowledge-incorporated dataset containing 1226 annotated images. The proposed automated workflow utilizing Grasshopper surpasses the Dynamo workflow in efficiency by 9 times. The BIM models have been integrated into a web-based digital twin platform, facilitating asset management and public engagement.</p
A novel study on hybrid physics-data-driven reduced-order modeling for aerodynamic load inversion under structural field uncertainties
No full textAerodynamic loads are crucial for the structural safety of aerospace vehicles. However, direct measurement or high-fidelity simulation remains challenging. In service, structural parameters are influenced by multi-source uncertainties, often characterized by limited samples and spatial variability. To address this, this paper proposes a novel aerodynamic load inversion method considering structural field uncertainties, which reconstructs load boundaries from sparse measurements. First, reduced-order modeling (ROM) is employed to project high-dimensional aerodynamic loads into low-dimensional subspaces based on spatial distribution features, transforming the load inversion problem into the prediction of feature coefficients. A physics-data-driven neural network (PDNN) is developed to predict these coefficients, integrating a data-driven module capturing the response-coefficient mapping relationship with a physics-based module that enforces physical consistency. This hybrid design improves the performance under small-sample and high-noise conditions. To model structural field uncertainties, the interval Karhunen-Loève decomposition (IKLD) is adopted to convert uncertain field into a finite set of uncertain parameters. An adaptive Kriging surrogate model (AKSM) is then used for uncertainty propagation, yielding the interval estimation of aerodynamic loads. The effectiveness of the proposed method is validated through numerical and experimental examples, demonstrating its high accuracy, robustness, and practical applicability, even under sparse measurements, noise interference and field uncertainties.</p
Some permutation pentanomials over finite fields of even characteristic
No full textIn a recent paper [30] Zhang et al. constructed 17 families of permutation pentanomials of the form x t + x r 1 (q − 1) + t + x r 2 (q − 1) + t + x r 3 (q − 1) + t + x r 4 (q − 1) + t over F q 2 where q = 2 m . In this paper for 14 of these 17 families we provide a simple explanation as to why they are permutations. We also extend these 14 families into three general classes of permutation pentanomials over F q 2 .</p
Quantitative characterization of deep defects in granular systems via inverse transient heat transfer analysis of active thermography
No full textAccurate quantification of subsurface structural defect and thermal anomalies in granular systems (GS) is essential for optimizing heat transfer in energy storage, thermal management, and construction materials. Conventional modalities such as X-ray CT, ultrasound, and GPR are often ineffective in highly scattering granular media, while active thermography (AT), though practical and non-destructive, has previously offered only qualitative or shallow defect insights. Here, we address the inverse transient heat transfer problem in GS by presenting a robust quantitative framework for simultaneous measurement of sizes and depths of defects at the millimeter scale. Our approach introduces a novel, spatially-informed figure of merit (FOM) to objectively benchmark contemporary thermogram denoising and transformation algorithms—including for defect contrast enhancement under high noise. The workflow integrates this optimized processing with segmentation-based feature extraction and systematically compared and optimized a suite of regression and machine learning approaches for mapping extracted features to defect geometry. Analytical model validation and finite element simulations, augmented with realistic synthetic noise, enable rigorous uncertainty analysis and robust performance assessment. The proposed method achieves mean relative errors of 3.6% for depth and 15.5% for radius, substantially outperforming conventional two-step direct thermogram regression (22.3% and 43.1% errors, respectively), and reliably quantifies defects with depth-to-diameter ratios up to 1.5 for both spherical and non-spherical geometries. This generalizable framework advances quantitative inverse analysis of transient heat transfer in granular systems, providing an effective tool for thermal quality control and optimization across energy, manufacturing, and infrastructure applications.<br/