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
Position Paper: Artificial Intelligence in Medical Image Analysis: Advances, Clinical Translation, and Emerging Frontiers
No full textOver the past five years, artificial intelligence (AI) has introduced new models and methods for addressing the challenges associated with the broader adoption of AI models and systems in medicine. This paper reviews recent advances in AI for medical image and video analysis, outlines emerging paradigms, highlights pathways for successful clinical translation, and provides recommendations for future work. Hybrid Convolutional Neural Network (CNN) Transformer architectures now deliver state-of-the-art results in segmentation, classification, reconstruction, synthesis, and registration. Foundation and generative AI models enable the use of transfer learning to smaller datasets with limited ground truth. Federated learning supports privacy-preserving collaboration across institutions. Explainable and trustworthy AI approaches have become essential to foster clinician trust, ensure regulatory compliance, and facilitate ethical deployment. Together, these developments pave the way for integrating AI into radiology, pathology, and wider healthcare workflows.</p
Quants and poets: two dimensions of MBA performance, aptitudes, and interests
No full textIntroduction: Research on MBA student performance typically relies on GPA as the primary indicator of success. However, business schools aim to develop future leaders for diverse career paths, which value multiple forms of performance. We examine whether performance is better understood as multidimensional, testing a longstanding distinction in MBA discourse between “poets” and “quants.” We also examine how different forms of admissions data (i.e. standardized test scores, undergraduate grades, stated interests, and pre-MBA experiences) predict distinct forms of success. Methods: We report results from two large-N studies using survey and archival data from an elite U.S. MBA program. Study 1 examines whether core course grades reflect multiple dimensions of academic performance and whether admissions-time aptitude measures differentially predict those dimensions. Study 2 replicates these findings using archival academic, extracurricular, and peer-evaluation records and extends the analysis to leadership outcomes. Confirmatory factor analysis and multivariate regression models are used across both studies. Results: Across both studies, MBA academic performance bifurcates into two weakly correlated dimensions: systematizing (quantitative, analytical success) and social (verbal, interpersonal success). These align with the popular MBA “poet vs. quant” distinction. Quantitative aptitude predicts quantitative academic performance, whereas verbal and writing aptitude predict social academic performance. Beyond grades, social performance is uniquely associated with leadership success, including both objective attainment (e.g., club leadership roles) and peer perceptions (e.g., assertiveness and inclusiveness). Student interests further differentiate outcomes: quant-oriented interests predict quantitative academic success but negatively predict leadership attainment, whereas poet-oriented interests positively predict leadership outcomes. Discussion: These findings demonstrate that MBA success is fundamentally multidimensional and that different admissions indicators predict different forms of performance, with implications for talent assessment, leadership development, and MBA admissions practice.</p
Programming Interfacial Polymerization: Machine Learning Unveils Quantitative Rational Design Rules for Microcapsules and Beyond
No full textInterfacial polymerization (IP) serves as a versatile platform technology for designing polymeric membranes, yet its extension to applications such as microencapsulation (MIP) remains hindered by empirical methodologies, largely due to the absence of quantitative rational design principles. Unlike separation membranes, which prioritize nanostructural control, MIP emphasizes encapsulation efficiency (EE%), rendering conventional membrane-derived theories and thermodynamic descriptors insufficient. In this work, we transcend these limitations by employing interpretable machine learning to program interfacial polymerization, thereby deciphering mechanism-informed quantitative design rules. Our data-driven platform integrates molecular thermodynamics, polymerization kinetics, and emulsion-stabilized interfacial parameters to identify previously overlooked descriptors governing microcapsule formation. We establish a predictive chemical–process–structure–performance relationship and demonstrate programmable control over key performances, including EE% (30%–95%), particle size (100–400 µm), and shell thickness-to-radius ratios (0.005–1) for diverse payloads spanning hydrophobic, hydrophilic, and highly reactive compounds such as toluene diisocyanate and amines. This work not only resolves long-standing challenges in understanding complex multiphase interactions in MIP but also establishes a new paradigm for the quantitative design of polymeric microcapsules, with broad implications for functional particles, catalytic microreactors, digital cells, and membranes.</p
Sex-specific toxicity targets of aristolochic acids: nephrotoxicity in males, hepatotoxicity in females
No full textAristolochic acids (AAs), derived from Aristolochia herbs, are well-known nephrotoxins, and emerging evidence suggests their potential role in the development of liver cancers. However, the specific organs most affected by AAs, particularly in relation to liver cancer, remain unclear. Considering the known sex differences in enzyme activities, we hypothesized that variations in AA metabolism may contribute to the kidney and liver toxicity associated with these compounds. Our analysis of DNA adducts in the kidneys and livers of mice treated with aristolochic acid I (AA-I) revealed that male mice exhibited over 2.5 times higher levels of DNA adducts in their kidney DNA compared to female mice. Conversely, female mice showed 1.5 times higher adduct levels in their liver DNA than their male counterparts. These findings indicate that AA exposure presents a sex-specific disease risk, with males being at greater risk for kidney disease and females for liver disease. Additionally, we observed similar concentration patterns of the metabolite aristolactam I (AL-I) and the activity of NQO1 enzymes in the respective organs. Further in vitro studies, involving the incubation of AA-I with liver and kidney homogenates, demonstrated significant differences in AL-I concentrations, mirroring the trends observed in the AA-DNA adduct and AL-I analyses of AA-I-exposed mice. Collectively, these results underscore the importance of sex differences in the enzymatic activity responsible for the metabolic activation of AAs, which is critical for understanding the differential nephrotoxic and hepatotoxic effects associated with these compounds.</p
Microwave-assisted catalytic upcycling of polyolefin wastes into hydrogen and carbon nanotubes over Pt-promoted Fe/Ni bimetallic catalysts
No full textThe rapid accumulation of plastic waste poses a critical environmental challenge and demands effective upcycling strategies to advance a circular economy. Microwave-assisted catalysis offers a low-energy, highly efficient alternative to conventional thermal processes; however, the principles guiding catalyst design and the underlying reaction mechanisms in this context remain insufficiently understood. Here, we report a microwave-assisted catalytic approach for converting polyolefin waste into high-value hydrogen (H2) and multi-walled carbon nanotubes (MWCNTs) using a Pt-promoted Fe/Ni bimetallic catalyst. The catalyst design leverages incorporation of a very small amount of Pt (e.g., 0.3 %) to substantially lower the oxygen vacancy formation energy, as revealed by experimental characterization and density functional theory (DFT) calculations, resulting in a 3.3-fold increase in strong acid site density. These modifications reduce the activation barriers for C–H and C–C bond cleavage, enabling efficient polyolefin conversion. Butane was employed as a model compound in DFT calculation to elucidate the polyethylene decomposition reaction mechanism, which proceeds via terminal C–H bond activation, formation of conjugated olefins, and subsequent deep dehydrogenation. Under microwave treatment, the Fe/Ni–0.3 %Pt catalyst achieved an H2 yield of 53.9 mmol/g and a selectivity of 90 % from low-density polyethylene (LDPE). The carbonaceous co-products were mainly high-quality MWCNTs exhibiting excellent electromagnetic interference (EMI) shielding performance. This integrated catalytic–microwave approach demonstrates a scalable route for transforming various polyolefin wastes into clean H2 and advanced carbon materials, offering a viable pathway for sustainable energy and material production within a circular economy framework.</p
Adaptive transfer learning with CPT sensor fusion for few-shot anomaly diagnosis in injection molding process
No full textInjection molding (IM) is a pivotal polymer processing technique that produces precision components across industries. Achieving consistent product quality remains a significant challenge due to inherent variations between material batches, frequent changes in operational conditions, and the critical need to detect and diagnose anomalies. These anomalies stem from deviations in cavity pressure, temperature, and material properties. Addressing these challenges, this paper proposes a few-shot anomaly diagnosis framework based on Capacitance-Pressure-Temperature (C-P-T) multi-sensor fusion and adaptive transfer learning, targeting 12 distinct anomaly classes. Based on C-P-T sensor data we design a symmetric dual-branch Convolutional Neural Network (CNN) architecture to extract domain-invariant features and a novel hybrid loss function to optimize few-shot learning. For cross-domain adaptation, a hierarchical transfer strategy freezes the first three convolutional layers of the source model while fine-tuning the last two layers and classifier, enabling rapid deployment from Acrylonitrile Butadiene Styrene (ABS) to Polypropylene (PP) material domains and different machine domains. Industrial validation demonstrates 98.93% accuracy in the source domain with only 25 samples per anomaly class, and 94.2% accuracy in target domains with just 5 samples. This framework provides high-precision intelligent process monitoring for flexible injection molding production.<br/
SceneLoom: Communicating Data with Scene Context
No full textIn data-driven storytelling contexts such as data journalism and data videos, data visualizations are often presented alongside real-world imagery to support narrative context. However, these visualizations and contextual images typically remain separated, limiting their combined narrative expressiveness and engagement. Achieving this is challenging due to the need for fine-grained alignment and creative ideation. To address this, we present SceneLoom, a Vision-Language Model (VLM)-powered system that facilitates the coordination of data visualization with real-world imagery based on narrative intents. Through a formative study, we investigated the design space of coordination relationships between data visualization and real-world scenes from the perspectives of visual alignment and semantic coherence. Guided by the derived design considerations, SceneLoom leverages VLMs to extract visual and semantic features from scene images and data visualization, and perform design mapping through a reasoning process that incorporates spatial organization, shape similarity, layout consistency, and semantic binding. The system generates a set of contextually expressive, image-driven design alternatives that achieve coherent alignments across visual, semantic, and data dimensions. Users can explore these alternatives, select preferred mappings, and further refine the design through interactive adjustments and animated transitions to support expressive data communication. A user study and an example gallery validate SceneLoom's effectiveness in inspiring creative design and facilitating design externalization.</p
Leveraging pre-trained models for kernel machines
No full textPre-training techniques have successfully promoted the training of neural networks. Since neural networks and kernel machines share similar properties, such as both learning the problems by the non-linear projection on features and both being capable of handling complex tasks, the idea of pre-training may also help kernel machines achieve promising training speed. However, existing pre-training-based kernel machine solvers show limited improvements on efficiency when the hyper-parameter varies. To effectively reduce the training cost, we propose a novel method that can make efficient use of pre-trained models to infer kernel machine models with different hyper-parameters. Our pre-training-based method is built on top of theoretical foundations. The difference between the model inferred based on pre-training and the optimal model is theoretically bounded by a constant. Experimental results show that our method can save an order of magnitude of training time compared with the existing approach while producing competitive accuracy.</p
Double-layer electromagnetic shielding materials with microcellular structure for strong absorption and low reflection
No full textGiven the rapid development in information and communication technologies, there is a pressing need for efficient, broadband, and lightweight shielding materials that emphasize electromagnetic (EM) wave absorption. Traditional single-layer material faces challenges such as limited absorption and secondary reflective pollution. To address the issues, we constructed a double-layer composite structure composed of a microcellular absorption layer comprised of multi-walled carbon nanotube (MWCNT)/flake iron powder (Fe)/poly (vinylidene fluoride) (PVDF) and a highly conductive solid MWCNT/PVDF (20/80 vol.%) layer. By controlling the composition of MWCNT/Fe in PVDF and the void fraction (VF) using supercritical carbon dioxide foaming, the impedance matching and the EM attenuation capability of the absorption layer were effectively adjusted. A reflection loss (RL) as low as −60 dB and an effective absorption bandwidth of 3.44 GHz were obtained for MWCNT/Fe/PVDF (5/10/85 vol.%) with a VF of 75 %. By combining the absorption and shielding layers, an average total shielding effectiveness of 41.75 dB and reflectivity of 0.12 across the X-band were achieved with a thickness of only 3 mm. The employment of metallic magnetic micro-particles improves impedance matching due to their high saturation magnetization and ease of integration with the dielectric material MWCNT. The combination of the microcellular structure and the highly conductive layer allowed for the decoupling of impedance matching, attenuation, and shielding, resulting in an “absorption-reflection-reabsorption” pathway for dissipating EM energy. The heterogeneously structured double-layer composite with excellent shielding efficiency and low reflection characteristics presents a promising strategy for developing next-generation high-performance, absorption-dominant EM interference shielding materials.</p
Cell Design for Practical High-Energy Lithium Metal Battery
No full textBatteries with Li metal anode show excellent promise for energy storage because of their high capacity, compatibility with Li-free cathodes (such as sulfur (S) and oxygen (O2)), and environmental friendliness. However, their coulombic efficiency is low, dendrite formation is unsafe, side reactions are serious, and the long-term service life is an issue. Therefore, taking the cell design for a practical high-energy lithium metal battery as an example, this review discusses four major challenges for the cell design of a high-energy lithium metal battery, which include the anode design, the separator design, the cathode design, and the external fields. Based on these contents, several key points for practical applications. of high-energy LMBs are outlined, including battery assembly and manufacturing, energy density, cell designs for LMBs with various electrolyte systems, economic perspective on the cell design, and battery management system.</p