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
Time-Lapse Video-Based Embryo Grading via Complementary Spatial-Temporal Pattern Mining
No full textArtificial intelligence has recently shown promise in automated embryo selection for In-Vitro Fertilization (IVF). However, current approaches either address partial embryo evaluation lacking holistic quality assessment or target clinical outcomes inevitably confounded by extra-embryonic factors, both limiting clinical utility. To bridge this gap, we propose a new task called Video-Based Embryo Grading - the first paradigm that directly utilizes full-length time-lapse monitoring (TLM) videos to predict embryologists’ overall quality assessments. To support this task, we curate a real-world clinical dataset comprising over 2,500 TLM videos, each annotated with a grading label indicating the overall quality of embryos. Grounded in clinical decision-making principles, we propose a Complementary Spatial-Temporal Pattern Mining (CoSTeM) framework that conceptually replicates embryologists’ evaluation process. The CoSTeM comprises two branches: (1) a morphological branch using a Mixture of Cross-Attentive Experts layer and a Temporal Selection Block to select discriminative local structural features, and (2) a morphokinetic branch employing a Temporal Transformer to model global developmental trajectories, synergistically integrating static and dynamic determinants for grading embryos. Extensive experimental results demonstrate the superiority of our design. This work provides a valuable methodological framework for AI-assisted embryo selection. The source code is available at https://github.com/RIL-Lab/CoSTeM.</p
Finite-volume converged ENO scheme with adaptive mesh resolution for steady-state compressible flow simulation
No full textIn Liang et al. [1], a novel perspective on the non-convergence property of high-order shock-capturing schemes and a new paradigm of finite-difference converged ENO (CENO) scheme considering the evolutionary information have been proposed. In this study, with the observation that compressible steady-state flow problems are typically characterized by the narrow shockwave regions of interest and other large regions with smooth flow scales, a new finite-volume CENO scheme with adaptive mesh resolution is innovatively proposed for the steady-state flow simulation, followed by three new concepts: (1) an improved converged stencil selection strategy with non-essential noise filtering is proposed to further improve the convergence property of CENO scheme for the finite-volume framework; (2) based on the idea that the inherent reconstruction procedure within the finite-volume framework not only provides the high-order reconstructed data at the cell interface for flux evaluation but also features the potential to examine the regularity of flow scales in the characteristic space, a new adaptation strategy with an explicit buffer region for the multi-resolution steady-state simulation is proposed; (3) a generalized approach for calculating the average residual of multi-resolution methods is proposed by constructing a tailored virtual “root” level through the coarsening-and-merging approach. A set of well-established challenging benchmark simulations involving strong shockwaves, contact discontinuities, and rarefaction waves demonstrates that the new finite-volume CENO scheme with adaptive mesh resolution not only features superior convergence properties but also can achieve a high data compression ratio, while producing a comparable solution with that from the uniform mesh at the highest effective resolution. Moreover, the present results serve as the state-of-the-art references for future numerical method development.</p
A correction symbol Granger causality approach to root cause diagnosis of non-stationary industrial processes
No full textDealing with non-stationary characteristics and revealing the symbol properties of causalities remain obstacles to performing fault root cause diagnosis (RCD) in industrial processes. Existing methods overlook these problems, leading to reduced RCD accuracy and a limited understanding of the dynamic characteristics of fault propagation. To address the above-mentioned problems, this paper proposes a correction symbol Granger causality (CSGC) approach. Firstly, to infer significant Granger causality (GC) from non-stationary processes, the cointegration relationships among variables are utilized to differentiate them and construct an error correction term, enabling the regression model to capture the long-term dependencies among fault variables better. Secondly, the self-explaining neural network is employed to further reveal the symbol properties of the fault variable from the fitted generalized coefficient matrix. Furthermore, two causal criteria of CSGC are established to test the significance level of GC and eliminate bidirectional causalities. Based on CSGC, a causal graph construction strategy containing information on symbol properties is developed for fault RCD and analysis of fault propagation paths. The analysis results of the numerical simulation system indicate that the various metrics of CSGC perform well, demonstrating its potential for accurate RCD and in-depth fault propagation analysis (Accuracy: 0.96, Sensitivity: 1.00, Specificity: 0.95, F1 score: 0.91). Finally, the effectiveness of RCD is validated through its application to the numerical simulation, the Tennessee Eastman process (TEP) and a real-world motor soft-foot fault.</p
Cine-CLIP: Reducing Cine-MRI Dimensionality with Temporal Variability for Left Ventricular Ejection Fraction Estimation
No full textCine cardiac magnetic resonance (cine-CMR) imaging is the gold-standard modality for assessing myocardial structure and function, including Left Ventricular Ejection Fraction (LVEF) measurement. However, cine-CMR presents challenges due to its high-dimensional 3D + time nature. To address this, we propose cine-CLIP, which employs standard deviation (std) mapping across time to reduce 4D data to 3D while preserving dynamic and spatial information. Through extensive experiments on publicly available UK Biobank and ACDC datasets, our method, cine-CLIP, achieves state-of-the-art (SOTA) LVEF prediction with a mean absolute error (MAE) of 2.523, outperforming other techniques. To assess generalizability, we further validate our method on the external Kaggle Data Science Bowl dataset, which follows a slightly different CMR acquisition protocol. Despite domain shifts, we achieved an MAE of 5.091, surpassing prior methods. These findings highlight Cine-CLIP’s ability to capture the high-dimensional complexity of cardiovascular disease from cine-CMR data. LVEF prediction serves as a proof of concept, demonstrating the model’s effectiveness in this task. However, this framework has the potential to be extended to other clinical metrics. The code is available at https://github.com/enriquealmar9/Cine-CLIP.</p
CoVis: Neural and LLM-Driven Multi-Turn Interactions for Conversational Text-to-Visualization Generation
No full textData visualization (DV) is widely used to communicate insights from large datasets. To lower the barrier to DV use, researchers have investigated automatic DV tasks like natural language question (NLQ) to visualization translation, formally called text-to-vis. However, text-to-vis assumes well-organized NLQs expressed in single sentences, which does not reflect real-world scenarios where complex DVs often require consecutive exchanges between users and systems. We introduce Conversational text-to-Visualization (CoVis), a novel task aimed at constructing DVs through multi-turn user-system interactions. We reconstruct and significantly extend Dial-NVBench through an enhanced multi-stage LLM-driven workflow with backwards planning, mandatory SQL execution validation, and intelligent error correction. The reconstructed dataset features improved executability and comprises dialogue sessions with diverse queries−ranging from dataset inquiries and data manipulation to visualization generation−reflecting real-world data visualization complexity. We investigate both neural-based and LLM-based approaches for this task. For the neural approach, we present comprehensive evaluations of MMCoVisNet, a multi-modal network which comprehends dialogue context and employs adaptive decoders: a text decoder for general responses, an SQL-form decoder for data queries (extended to support multi-table joins), and a DV-form decoder for visualizations. For the LLM-based approach, we develop CoVisLLM with a novel query decomposition and reconstruction framework featuring five systematically designed tasks that break down complex reasoning into structured intermediate representations. Comparative evaluation shows improvements over strong baselines for both MMCoVisNet and CoVisLLM.</p
Speech-to-Visualization: Toward End-to-End Speech-Driven Data Visualization Generation from Natural Language Questions
No full textData visualization (DV) has evolved rapidly, transforming intricate datasets into accessible visual re presentations. However, the intricate grammar of DV languages, such as Vega-Lite, presents a substantial barrier for beginners and users without technical backgrounds. To address this challenge, extensive research has focused on developing models that can translate natural language questions (NLQs) into DV languages, a process formally known as text-to-visualization in the field. With the recent development of speech-related technologies, particularly Acoustic Speech Recognition (ASR), voice-based interaction has become a growing trend in real-world applications. In this paper, we introduce speech-to-vis, a novel task that translates speech-form NLQs into data visualizations. To address the scarcity of relevant datasets, we present SpeechNVBench, the first manually annotated dataset specifically designed for this field. Our research reveals that the intuitive cascaded approach (i.e., ASR followed by text-to-vis) suffers from error propagation issues, where small errors in earlier stages lead to larger errors in subsequent stages. In response, we introduce SpeechVisNet, the first end-to-end neural architecture that directly translates speech-form NLQs into DVs. SpeechVisNet incorporates advanced structures like a DV-aware decoder to ensure reliable output. Furthermore, to mitigate the modality gap between speech-modality questions and text-modality data schema, we explore bridging techniques to align them. Experimentation on our proposed dataset demonstrates SpeechVisNet’s competitive edge against various strong baselines. This work aims to drive innovation in human-machine interfaces, enhancing the efficiency and accessibility of DV tools across various domains.</p
A fully coupled THMC-MPM framework for modeling phase transition and large deformation in methane hydrate-bearing sediment
No full textMethane hydrate-bearing sediment (MHBS) is a multiphase granular system characterized by complex thermo-hydro-mechanical–chemical (THMC) interactions involving phase transitions and large deformation behavior. Hydrate dissociation weakens sediment strength, potentially initiating geohazards such as submarine landslides. Simultaneously, large deformations in MHBS alter the sediment's state, influencing hydrate reaction kinetics. Despite recent progress, modeling the coupled processes of hydrate dissociation and large deformation in MHBS remains a significant challenge. This study develops a THMC-coupled material point method (MPM) framework to simulate the pre- to post-failure behavior of MHBS associated with hydrate dissociation. The framework incorporates three key advancements: (i) a six-field governing equation integrated with the Kim–Bishnoi hydrate reaction model to resolve dynamic phase transitions, multiphase interactions, and large deformations; (ii) a strain-softening Mohr–Coulomb model with hydrate saturation-dependent strength to capture sediment mechanical degradation; and (iii) a hybrid explicit–implicit time integration scheme designed to enhance computational efficiency for systems with low permeability and high reaction rates. The framework is validated against Masuda's hydrate dissociation experiment and an extended Terzaghi consolidation benchmark, before being applied to simulate biaxial compression tests and hydrate dissociation-triggered slope failures. We reveal that (1) shear dilation generates negative excess pore pressure in undrained conditions, triggering hydrate dissociation within the shear bands; (2) shear heating resulting from rapid, large deformation promotes hydrate dissociation, exacerbating sediment softening; and (3) sediment strength degradation, hydrothermal variations, slope geometry, and other factors collectively shape the dynamic progression of retrogressive failures in MHBS. This work provides a powerful framework for modeling hydrate-related granular mechanics and geohazards.</p
Origin and Deployment of Qing Military Officers by Province, 1786–1908 CE
No full textWe investigate changes over time in the distribution by province of current post and province of origin for Qing military officers from the late eighteenth century to the beginning of the nineteenth century. During this period, the Qing faced a variety of military challenges, including domestic conflicts and foreign incursions. The most important was the Taiping Rebellion (1851–1864), which is already known to have led to large changes in the composition of the Qing military leadership. In turn, senior Hunan-origin military officers leveraged their networks to dominate officer appointments in the coastal provinces. We examine how the Taiping Rebellion, the First Opium War and other crises affected the allocation of officers between provinces, and the recruitment of officers from different provinces. For the analysis, we use the quarterly rosters of military officers Zhongshu beilan, which have been transcribed into a database as part of the China Government Employee Dataset-Qing Jinshenlu (CGED-Q JSL). We show that the allocation of officers by province did not change during the First Opium War, but changed dramatically after the Taiping Rebellion, with a substantial increase in the share of officers allocated to the southeastern coastal provinces, reflecting heightened importance of maritime defense. We also show that there were two phases to the increase in the share of Hunan-origin officers, one at the end of the eighteenth century, and the other, better-known one following the Taiping Rebellion. Finally, we show that exceptions to the rule of avoidance in the appointment of senior military officers became more common for all types of officers from the eighteenth century to the middle of the nineteenth century, and that afterward, low-, mid-, and high-level officers followed different trajectories. We conclude with an assessment of the implications of our findings for our understanding of the Qing in the nineteenth century.</p
Proprioception and vision relationship in aimed movement with restricted and reversed vision
No full textHuman movement relies on visual and proprioceptive inputs, and the movement can be hampered when either input provides limited or distorted information, as in robotic surgery, virtual reality, or spaceflight. This study examines how proprioception influences performance in aimed movement tasks under restricted and inverted visual conditions. An experiment with 24 participants showed slower movements and increased information processing demands with incompatible visual feedback. The difference was diminished when the proprioceptive variability was controlled through ANCOVA. The findings allude to the necessity of considering individual differences in proprioceptive ability when evaluating performance across different visual conditions. These results can help develop training protocols and design interfaces when visual and proprioception sensory feedback are incompatible. Also, continuous visual feedback of hand movement may not be necessary. Instead, a correctly sized visual window of the target is sufficient for aimed movement tasks. This result has practical implications, especially when optimizing VR renders.</p
Power Consumption Forecasting of Spacecraft Based on Adaptive Frequency-Domain Pruning-Enhanced Transformer
No full textForecasting the power consumption of the spacecraft is critical for optimizing its lifespan and task allocation. However, the complex electromagnetic environment of outer space introduces unavoidable noise into the collected electrical signals. Moreover, the various subsystems of a multipower spacecraft are affected differently by internal and external noise, making it challenging for the existing methods to effectively capture the features of long-term power consumption sequences. We propose adaptive frequency-pruning-enhanced (AFPE)-iTransformer, a robust time-series forecasting model designed for spacecraft telemetry forecasting under noise and long-range dependency conditions. The model combines three key components: Legendre memory projection for historical compression, adaptive top-k frequency pruning for per-channel denoising, and an improved inverted transformer for cross-subsystem attention. Evaluated on three years of Mars Express (MEX) data, our method consistently outperforms the state-of-the-art baselines in both within-year and cross-year forecasting. It also achieves competitive efficiency, with fast model load time and moderate parameter size. While focused on power forecasting, the model’s modular design supports broader applications in telemetry and industrial forecasting. Model code and configurations are open-sourced for reproducibility.</p