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
Celebrating funerals as weddings: a computational text study of Chinese foundations’ COVID-19 sentiment responses
No full textCurrent literature has increasingly recognized the significance of cultural responses to disasters. While it largely focuses on the roles of the Chinese state or individuals, other actors’ involvement in producing disaster narratives and emotions remains under-explored. This study examines civic organizations’ meso-level participation in constructing the “celebrating funerals as weddings” (sangshixiban 丧事喜办) COVID-19 rhetoric. Using machine learning and statistical analysis, it analyzes an original WeChat dataset of 193 Chinese foundations and their organizational data. We found that foundations displayed predominantly positive sentiments about the pandemic on social media, with grassroots entities exhibiting significantly more positivity than their affluent, government-backed counterparts. Grassroots ones also attributed more affirmative values to the party-state, healthcare workers, and women while avoiding contentious entities like Zhang Wenhong, World Health Organization (WHO), and the United States. Despite their important presence in material disaster relief, grassroots foundations largely upheld the state’s discursive authority, sidestepping dissenting voices during the pandemic’s early height.</p
Modeling and dynamic analysis of belt drive systems using a harmonic balance and alternating frequency/time domain method
No full textPurpose – To guarantee smooth transmission and develop digital twins for status monitoring of belt drive systems (BDSs), this research aims to develop a physical model and perform dynamic analysis for BDSs. Design/methodology/approach – A mathematical model of a three-pulley BDS is first developed in both the time and frequency domains, considering the hysteresis characteristics of the automatic tensioner. Then, a harmonic balance and alternating frequency/time domain (HB-AFT) method is proposed to calculate the dynamic characteristics of BDSs. For dynamic analysis, the frequency response function (FRF) and energy dissipation characteristics of automatic tensioners, as well as their influencing factors, are calculated and analyzed. Finally, the digital twins development and fault diagnosis methods of BDSs are proposed using the dynamic analysis results. Findings – The dynamic characteristics of the BDS calculated by the HB-AFT method and traditional numerical methods are compared, and the comparison results show that the HB-AFT method has a calculation speed more than 40 times faster than the numerical iteration method. The FRF and energy dissipation characteristics of tensioners are related to the torsional vibration amplitude of the driving pulley and the sliding frictional torque of the tensioner, and the fault in tensioners can be detected by their oscillation amplitude calculated by digital twins. Originality/value – The proposed HB-AFT method can be used to calculate the dynamic characteristics of BDSs with high accuracy and efficiency, while considering the hysteresis characteristics of the tensioner. It can also be used for status monitoring and fault diagnosis in digital twins of BDSs.</p
Survey on AI-Enabled Computer Vision Technologies and Applications for Space Robotic Missions
No full textThis survey provides a comprehensive overview of recent advancements and challenges in Artificial Intelligence (AI)-enabled computer vision (CV) techniques for space robotic missions, spanning critical phases such as Entry, Descent, and Landing (EDL), orbital operations, and planetary surface exploration. Emphasis is placed on deep-learning–based approaches for image classification, object detection, semantic segmentation, relative pose estimation, and feature matching. State-of-the-art methods in terrain-relative navigation, crater-based or rock-feature matching, and pose estimation for uncooperative targets are highlighted, illustrating the progress achieved through hybrid pipelines combining deep neural networks with classical geometry. The paper also critically evaluates publicly available orbital and planetary data sets—along with the increasing role of synthetic data—for developing and benchmarking CV algorithms under strict resource limitations and harsh environmental conditions. Despite demonstrated success in tasks like autonomous landing, debris removal, and rover navigation, current solutions face significant hurdles. These include computational constraints on onboard hardware, insufficient coverage of planetary conditions in existing data sets, and limited adaptability to dynamically changing environments. To address these shortcomings, research must prioritize lightweight neural architectures, advanced synthetic data generation, adaptive or incremental learning, and robust multisensor fusion. By integrating these strategies, AI-based CV systems can advance autonomy, precision, and resilience in future space missions.</p
Single-Crystalline Borate Covalent Organic Frameworks for Solid-State Lithium Metal Batteries
No full textTo advance lithium metal batteries, novel solid-state electrolytes are crucial. Covalent Organic Frameworks (COFs) are promising due to their crystalline, porous structure, light composition, strong bonds, high surface area, and stability. COFs can be engineered for enhanced ion conduction, with their porosity and ion-functional groups enabling fast ion movement and uniform lithium deposition. We synthesized a single-crystalline 3D borate COF (B-COF), achieving an ionic conductivity of 8.1 mS cm−1 at room temperature and a lithium-ion transference number of 0.98 in a quasi-solid-state. In symmetric cells, B-COF supported stable lithium deposition/stripping for 2000 h, suppressing dendrite formation. Full cells with LiFePO4 cathodes cycled stably at 0.5C, delivering 147 mAh g−1 initial capacity, 91.8% retention, and 99.98% Coulombic efficiency over 600 cycles. These results highlight B-COF's potential as a high-performance solid electrolyte for lithium metal batteries.</p
Sensing for Free: Learn to Localize More Sources than Antennas without Pilots
No full textIntegrated sensing and communication (ISAC) represents a key paradigm for future wireless networks. However, existing approaches often require waveform modifications, dedicated pilots, or additional overhead that complicates standards integration. We propose “sensing for free”—performing multi-source localization without pilots by reusing random and unknown uplink data symbols, where sensing happens simultaneously with data transmission, making it directly compatible with the 3GPP 5G NR and 6G specifications. With the ever-increasing number of devices in dense 6G networks, this approach becomes particularly compelling when combined with sparse arrays, which can localize a much larger number of sources compared to uniform arrays through the enlarged virtual array. However, existing pilot-free multi-source localization algorithms for sparse arrays have numerous drawbacks. They mostly first reconstruct an extended covariance matrix and then apply subspace methods, which incur prohibitive cubic complexity while being limited to second-order statistics. Performance degrades under non-Gaussian modulated data symbols in cellular wireless networks as the higher-order statistics that could further enhance the localization capability remain unexploited. We address these challenges with an attention-only transformer that directly processes raw signal snapshots for grid-less end-to-end direction-of-arrival (DOA) estimation. The model efficiently captures higher-order statistics while being permutation-invariant and adaptive to varying numbers of snapshots. Our algorithm greatly outperforms state-of-the-art artificial intelligence (AI)-based benchmarks with over 30× reduction in parameters and runtime, and enjoys excellent generalization under practical mismatches. In addition, it can effectively handle multipath propagation and mixed modulation types. Beyond localization, our algorithm can also enhance multi-user MIMO beam training through angular reciprocity. The estimated DOAs in the uplink data transmission stage can significantly prune downlink beam sweeping candidates and enhance system throughput via sensing-assisted beam management. Overall, this work demonstrates how reusing existing random data payloads for sensing can enhance both multi-source localization and beam management, two key AI-for-communication use cases in 3GPP efforts towards 6G.</p
Integrating equivariant architectures and charge supervision for data-efficient molecular property prediction
No full textUnderstanding and predicting molecular properties remains a central challenge in scientific machine learning, especially when training data are limited or task-specific supervision is scarce. We introduce the molecular equivariant transformer (MET), a symmetry-aware pretraining framework that leverages quantum-derived atomic charge distributions to guide molecular representation learning. MET combines an equivariant graph neural network (EGNN) with a transformer architecture to extract physically meaningful features from three-dimensional molecular geometries. Unlike previous models that rely purely on structural inputs or handcrafted descriptors, MET is pretrained to predict atomic partial charges, which are quantities grounded in quantum chemistry. This enables MET to capture essential electronic information without requiring downstream labels. We show that this pretraining scheme improves performance across diverse molecular property prediction tasks, particularly in low-data regimes. Analyses of the learned representations reveal chemically interpretable structure–property relationships, including the emergence of functional group patterns and smooth alignment with molecular dipoles. Ablation studies confirm that the EGNN encoder plays a crucial role in capturing transferable spatial features, while the transformer layers adapt these features to specific prediction tasks. This architecture draws direct analogies to quantum mechanical basis transformations, where MET learns to transition from coordinate-based to electron-based representations in a symmetry-preserving manner. By integrating domain knowledge with modern deep learning techniques, MET offers a unified and interpretable framework for data-efficient molecular modeling, with broad applications in computational chemistry, drug discovery, and materials science.</p
The association between depressive symptom severity and metabolic disturbances in major depressive and bipolar disorders: A systematic review and meta-analysis
No full textBackground Persons with depression are differentially affected by metabolic alterations, notably, insulin resistance and dyslipidemia. Metabolic alterations affect acute pharmacotherapy response and predispose risk for cardiovascular diseases. We aimed to extend knowledge pertaining to the depression-metabolic alteration association by evaluating whether depressive symptom severity moderates the association. Methods We conducted a systematic search of PubMed, Ovid and Scopus from inception to May 2025. Two reviewers (S.W. and G.H.L.) independently screened the identified studies. Studies were included if they enrolled adults with depression and reported on at least one metabolic parameter (i.e., fasting glucose, insulin, lipid panels). Standardized mean differences of metabolic parameters were pooled across studies. Results We identified 28 studies for inclusion. Persons with depression exhibited higher fasting glucose (SMD = 0.30, 95 % CI [0.12, 0.48]) and dyslipidemia [i.e., trends of increased low-density lipoprotein (SMD = 0.21, 95 % CI [−0.03, 0.44]) and lower high-density lipoprotein (SMD = −0.72, 95 % CI [−1.41, −0.03])]. Measures of insulin resistance were positively associated with anhedonia severity, sleep disturbances, and suicidal ideation. Limitations Between-study methodological differences, including study design and sociodemographics, affects the synthesis of overall trends. Conclusion Herein, we identify an association between depressive symptom severity and dysglycemia, dyslipidemia and insulin resistance. The results augment the conceptual framework implicating metabolic disturbances in depression pathophysiology and indirectly support testing that therapeutics currently in development in the treatment of depression (e.g., GLP-1 receptor agonists) may exhibit differential efficacy as a function of illness severity.</p
The Ricci–DeTurck flow from an initial metric with a Morrey-type integrability condition
No full textWe study the short-time existence theory of Ricci–DeTurck flow starting from rough metrics that satisfy a Morrey-type integrability condition. Using the rough existence theory, we show the preservation and improvement of distributional scalar curvature lower bounds provided the singular set for such metrics is not too large. As an application, we use Ricci flow smoothing to study the removable singularity related to scalar curvature. Our result supplements those of Jiang, Sheng and Zhang
General synthetic iterative scheme for fast solving coupled electron–phonon Boltzmann equations
No full textUnderstanding and controlling electron–phonon interactions is essential for optimizing thermal performance in microelectronic and thermoelectric devices, where strong non-equilibrium effects span multiple length and time scales. The coupled electron–phonon Boltzmann equations offer an insightful framework for modeling such transport, but their high dimensionality and stiffness pose significant computational challenges. Conventional iterative solvers typically suffer from excessive numerical dissipation and slow convergence in diffusive regimes. In this work, we develop a general synthetic iterative scheme (GSIS) that significantly accelerates the solution of coupled electron–phonon Boltzmann equations. The key innovation is the formulation of macroscopic synthetic equations that effectively capture diffusion-limit behavior, while precisely incorporating non-equilibrium corrections extracted from the kinetic level. During iterations, the kinetic solver provides high-order moment closures, while the macroscopic equations update the driving fields, ensuring rapid convergence. This strategic coupling across scales facilitates efficient global information exchange and robust multiscale resolution. Fourier analysis and numerical benchmarks show that GSIS can reduce computational time by up to three orders of magnitude compared to the conventional iterative scheme. Implemented with a high-order discontinuous Galerkin method, the GSIS is applicable to complex geometries and extendable to nonlinear and non-gray transport models.<br/
ICG-001 Provides Cardioprotection Against Doxorubicin-Induced Cardiotoxicity and Enhances Cancer Cytotoxicity
No full textDoxorubicin is effective against cancer but can cause doxorubicin-induced cardiotoxicity (DCT). Drug discovery efforts against DCT are hampered by the need to balance cardioprotection and cancer control. This study demonstrates that ICG-001 suppressed DCT in patient-derived human induced pluripotent stem cell–derived cardiomyocytes in vitro and in mice in vivo, comparable to conventional treatment, dexrazoxane. Unlike dexrazoxane, ICG-001 was cytotoxic to cancer cells. Mechanistically, ICG-001 protected the mitochondria in cardiomyocytes via DPR1 inhibition, but suppressed cancer by repressing Wnt signaling. These dual mechanisms underscore the potential of ICG-001 as an adjunct treatment to doxorubicin to improve its safety and efficacy.</p