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
IUTF Dataset: Enabling Cross-Border Resource for Analysing the Impact of Rainfall on Urban Transportation
No full textUnderstanding the impact of extreme weather, particularly flooding, on urban transportation systems is critical for enhancing city resilience and traffic management. However, research and policy development are often hampered by a lack of datasets that comprehensively integrate detailed traffic dynamics, high-resolution weather information, and road network topology across multiple diverse urban environments. To address this significant gap, we present the Integrated Urban Traffic-Flood (IUTF) dataset. This open-access resource covers 40 major cities across Europe, North America, and Asia, including 21,739 sensors. The IUTF dataset uniquely combines (i) high-resolution traffic parameters derived from over 21,700 sensors (with raw data typically at 5-minute intervals, harmonised to hourly); (ii) detailed hourly precipitation data from ERA5 reanalysis, spatially aligned with (iii) the underlying road network topology for over 1 million road segments, processed from OpenStreetMap. This meticulously curated and validated dataset, created through a novel spatio-temporal harmonisation framework, enables unprecedented, cross-border analysis of weather impacts on urban mobility. It provides a foundational data resource to support applications in traffic flow prediction, infrastructure planning, and the future development of quantitative resilience models.</p
The missense mutation Y65C in PQBP1 causes microcephaly and cognitive deficits through a combination of partial loss-of-function and gain-of-function effects
No full textThe missense mutation Y65C in polyglutamine-binding protein 1 (PQBP1) is associated with Renpenning syndrome, characterized by X-linked intellectual disability and microcephaly. However, the pathogenic mechanism underlying the microcephaly induced by the Y65C mutation remains unclear. In this study, we generated Pqbp1Y65C/Y knock-in male mice and discovered that the Y65C mutation impairs the proliferation of apical progenitors and their subsequent transition to basal progenitors, resulting in microcephaly and cognitive deficits like those observed in Renpenning syndrome patients. This Y65C substitution induces PQBP1 misfolding, which reduces PQBP1 protein levels and consequently impedes apical progenitor proliferation. Unexpectedly, the Y65C mutation also induces a gain-of-function that interferes with the transition from apical to basal progenitors by enhancing interactions with the core components of the mRNA 3’ end processing machinery, thereby preserving proliferative alternative polyadenylation (APA) profiles. Our study demonstrates that a combination of loss-of-function and gain-of-function contributes to the microcephaly caused by the Y65C mutation.</p
Comparative analysis in sensitivity of PM<sub>2.5</sub> mass to ammonia and nitrate availability between Hong Kong and Shanghai reveals comparative importance of chemistry and meteorology
No full textReductions in sulfate over recent years have increased the relative contributions of nitrate and ammonium to PM2.5 in Chinese megacities, shifting the focus of further PM abatement toward these semi-volatile species. This shift in focus makes understanding gas-particle partitioning, which is governed by aerosol pH and temperature, central to predicting their particle-phase contributions. Using hourly measurements of water-soluble inorganic ions (WSIIs) and related gases in Hong Kong and Shanghai, we constrained aerosol pH and aerosol water content (AWC) with ISORROPIA-II and evaluated PM2.5 mass sensitivity to total nitrate (TNO3 = HNO3 + NO3−) and total ammonia (TNH3 = NH3 + NH4+). In Hong Kong, sulfate dominated the WSIIs, whereas nitrate was dominant in Shanghai. The seasonally lower temperature (≈ 6–9 °C) in Shanghai, combined with a lower sulfate fraction, contributed to an aerosol pH approximately one unit higher than that in Hong Kong. The combination of higher pH and lower temperature in Shanghai favored particulate nitrate formation (high ε(NO3−)), increasing PM sensitivity to TNO3. Our quantitative sensitivity analysis showed that cutting TNO3 in Shanghai was as effective as cutting sulfate for PM2.5 reduction, while TNH3 controls required reductions of >40% to become effective. Consistent with its distinct chemical and meteorological regime, PM2.5 in Hong Kong was co-sensitive to both TNH3 and TNO3, indicating that a synergistic control strategy is optimal. Meteorology modulated these sensitivities primarily by altering aerosol pH and partitioning. Temperature exerted the strongest influence by directly controlling the thermodynamic equilibrium. High relative humidity favored greater aerosol water content, elevating aerosol pH and further promoting nitrate partitioning. Chemical composition, notably the sulfate and nitrate levels, also played a decisive role by setting the initial chemical regime. This contrast demonstrates that the lower temperatures and higher aerosol pH in Shanghai amplifies PM2.5 sensitivity to TNO3, warranting prioritized NOx control. In warmer Hong Kong, the lower aerosol pH and dominant role of sulfate result in co-sensitivity to both precursors, necessitating coordinated abatement. Thus, effective PM2.5 mitigation requires strategies tailored to local chemical-meteorological regimes.</p
Reliable building inventory imputation for regional-scale risk assessment: An uncertainty-guided framework using spatially-enhanced Transformers
No full textRegional-scale risk assessments require complete building inventories, yet missing data is a pervasive problem that undermines exposure models and their downstream risk analysis. This study addresses how to impute missing building attributes by explicitly modeling spatial dependencies in a scalable and expressive manner, while simultaneously quantifying the epistemic uncertainty from imputation, moving beyond the limitations of deterministic point estimates. We propose a novel framework that leverages a spatially-aware Feature Tokenization Transformer to impute missing values, then quantifies imputation uncertainty using Monte Carlo Dropout, and progressively refines estimates in an iterative process. On large-scale datasets from San Francisco, our proposed method consistently and significantly reduces imputation error, in some cases nearly halving it over the best performing baseline. It enhances the reliability of regional urban imputation tasks, enabling robust and confident decision-making.</p
Process-induced thermal conductivity degradation in Cu micro pads for 3D-IC hybrid bonding revealed by spatially resolved frequency-domain thermoreflectance
No full textWith three-dimensional integrated circuits (3D ICs) advancing toward mainstream adoption, Cu interconnects in through‑silicon vias and hybrid-bonding pads become increasingly critical for both signal transmission and heat dissipation from buried layers where Cu microstructure governs thermal transport. Modern 3D IC thermal simulations often assume bulk Cu thermal conductivity ( k Cu ≈ 400 W m−1 K−1), neglecting grain-boundaries, interfacial defects, and size effects in microscale features that significantly depress k Cu, underestimating device temperatures and reliability risks. Despite the spatial resolvability, conventional frequency-domain thermoreflectance (FDTR) models fail for laterally confined pads in dielectrics (SiO2) recesses when the thermal diffusion length approaches the pad radius. Here, we systematically investigate thermal transport while simultaneously probing geometries of Cu-SiO2 hybrid-bonding microstructures fabricated via industry-standard damascene processes by integrating FDTR with finite-element forward modeling and iterative numerical inversion. This framework captures lateral confinement of heat transfer and realistic structural defects (e.g., sidewall gaps and base voids) in microscale Cu pads. Benchmarking against well-established analytical models for bulk and blanket-film references validates the approach to within 1.5%, while sensitivity-guided fitting enables simultaneous extraction of k Cu, pad thickness, and Au/Cu interfacial conductance. Critically, confined Cu pad (5 μm-radius) in SiO2 recesses exhibit in-plane k Cu of only 160–220 W m−1 K−1, 30–50% lower than blanket films of similar thickness (∼315 W m−1 K−1) fabricated by the same process. Electron backscatter diffraction (EBSD) and cross-sectional scanning electron microscopy (SEM) reveal nanocrystalline grains (∼137 nm) and interfacial voids that intensify electron-grain boundary scattering with abnormally enhanced boundary reflection probability, accounting for the reduced k Cu. We further validate the framework's universality via time-domain thermoreflectance (TDTR) simulations and confirm the negligible impact of non-equilibrium carrier dynamics using a two-temperature model, justifying computationally efficient single-temperature inversion for industrial monitoring. As the first coupled experimental-numerical framework applied to realistic hybrid bonding structures, this methodology bridges the gap between blanket-film metrology and device-level reality, providing essential, experimentally grounded inputs for predictive 3D-IC thermal design.</p
Some uncommon observations of suspending particulates in the air in Hong Kong by ground-based remote-sensing meteorological instruments
No full textSuspending particulates such as volcanic ash, sand/dust and smoke from fire are not commonly observed in Hong Kong. There are some recent events with the setting in of such particulates and their observations by the ground-based remote sensing meteorological instruments are documented in this paper. The equipment in use includes an aerosol LIDAR, Doppler LIDARs and dual polarization weather radars. They are found to provide new insights into the observations of these suspending particulates in Hong Kong, not observed in the territory before and generally consistent with the observations of similar phenomena in other parts of the world. In particular, the equipment provides clear signature about the set-in, dispersion and, if possible, deposition of the particulates, and such observations are found to be in line with the meteorological forecasts by numerical weather prediction models, including a dispersion model whose results are described here. This paper is considered to be useful as reference for weather monitoring of such uncommon particulates in this part of the world.</p
A smart predict-then-optimize framework for vehicle rebalancing problem
No full textMatching the imbalanced supply and demand through vehicle rebalancing is critical for enhancing the operational efficiency of ride-hailing platforms. However, the traditional two-stage Predict-then-Optimize (PO) framework suffers from a mismatch between the loss function of the upstream prediction model and the objective function used in downstream decision-making. To tackle this challenge, we propose a Smart Predict-then-Optimize (SPO) framework, in which the prediction model is trained to directly minimize decision loss. Firstly, we formulate the regional-level vehicle rebalancing problem as a mixed integer linear programming (MILP) model, aiming to maximize the Gross Merchandise Volume (GMV) of the ride-hailing platform. After that, the Spatial and Temporal Identity (STID) model is employed to predict future demand and supply. Instead of training the prediction model by minimizing fitting error, we adopt a decision-focused loss function determined by the solution of the optimization model. Considering that uncertain parameters appear in the constraints, we develop a penalty-augmented loss function along with a corresponding solution adjustment method. Moreover, we propose a perturbation-based method to address the challenge of gradient backpropagation through the non-differentiable optimization layer, which enables the gradients of the decision loss to be obtained via zeroth-order approximation. The theoretical properties are checked, showing that the method can yield an unbiased approximation of the gradient. We conduct extensive experiments on a real-world dataset from Didi Chuxing, including both numerical studies and simulation experiments. The results show that the proposed SPO framework improves the average GMV by 2.19% compared to rule-based rebalancing and by 0.28% compared to the PO strategy. In particular, the prediction model trained by the SPO method can learn the utility of each region, enabling more effective vehicle rebalancing by dispatching drivers from low-utility origins to high-utility destinations.</p
Coupled grain-pore fabric evolutions in sheared granular materials: Anisotropy lagging and geometric emergence
No full textThe morphological evolution of pore spaces is a critical yet poorly quantified microstructural determinant of the macroscopic mechanical and hydraulic behavior of granular materials. While the anisotropy of the grain contact network (Fc) is known to dictate material response, the concurrent evolution of pore space anisotropy (Fp) and its coupling with Fc remains inadequately understood. This study employs Minkowski moment tensor analysis within a Discrete Element Method (DEM) framework to bridge this gap. We systematically investigate dense and loose, monodisperse and polydisperse assemblies under cyclic triaxial loading to quantify the dynamic coupling between Fc and Fp. We demonstrate a moderate to strong correlation between Fc and Fp, with a systematic lag in the response of Fp attributed to hierarchical geometric emergence across scales. This lag is constrained by particle-scale free-volume reorganization and its kinematic compatibility with particle motion. Additionally, key pore-scale metrics, including inverse Voronoi cell fractions (ϕv−1), pore-scale porosity (ϕp), and pore shape anisotropy β^, are well described by gamma distributions across all packing densities and strain levels. Notably, the scaled ϕv−1 follows a k-gamma distribution, providing a statistically consistent descriptor for volume fluctuations. A strong correlation is also observed between the average pore shape factor (\β\avg) and global porosity, suggesting that \β\avg serves as a geometry-based descriptor linking collective pore deformation to packing density. These findings underscore the utility of the Minkowski tensor approach in capturing 3D fabric evolution and explicitly linking pore- and grain-scale interactions. The quantitative relationships and statistical descriptors presented here provide a new foundation for enhancing constitutive models in geotechnics and powder technology, offering insights relevant to future investigations into permeability evolution and shear band formation.</p
AFedLF: Adaptive Layer Freezing of Foundation Models in Heterogeneous Federated Learning
No full textThe rise of pre-trained foundation models (FMs) has popularized the trend of fine-tuning FMs to fit downstream tasks, while Federated Learning (FL) has become the de-facto approach for training distributed data with privacy-preservation. However, fine-tuning FMs in FL faces overwhelming overheads due to its bulky nature. While freezing parameters in FM have the potential to accelerate FL training, existing freezing strategies statically freeze parameters on specified or already converged layers, incur severe accuracy degradation, and resource-inefficiency in heterogeneous environments. In this paper, we propose AFedLF, an adaptive freezing framework for FM in FL, to accelerate its wall-clock time for convergence without losing its final accuracy. However, this poses great challenges, as different freezing strategies lead to different accuracy gains and time overheads, while unfreezing more layers may bring marginal accuracy gains but significant time overheads. To address this challenge, AFedLF mathematically establishes a correlation between the freezing strategy and the accuracy gain and time overhead, and allocates adaptive freezing strategies to clients, based on our insight that unfreezing more layers on devices with strong computation and communication capabilities helps improve resource efficiency. Besides, AFedLF incorporates our well-designed intermediate result caching scheme with constant approximation ratios utilizing the limited storage capacity on mobile devices to cache intermediate results to skip forward propagation, further saving wall-clock time. Finally, we implemented AFedLF using an open-source FL benchmark, and extensive trace-driven experimental results showed that AFedLF accelerates wall-clock time by up to 6.1× compared to state-of-the-art solutions, without sacrificing accuracy.</p
