Selected work

Representative Papers

Designing PULSE: A Realtime Annotation Tool to Support Simulation Debriefing

Aug 10, 2026

This study addresses the challenges of high instructor workload and low observer engagement in simulation-based education, which often compromise debriefing effectiveness. In collaboration with nursing educators, the authors developed PULSE, a real-time annotation tool that introduces a novel student-led annotation mechanism. PULSE enables observing students to actively document their observations and reflections during simulations, thereby enhancing the structure, depth, and interactivity of debriefings. Grounded in human-centered co-design and informed by the DASH (Debriefing Assessment for Simulation in Healthcare) framework alongside qualitative interviews, the study systematically developed and validated an interactive debriefing support mechanism. Preliminary results demonstrate that PULSE significantly improves DASH scores (p = 0.027, Cohen’s d = 2.05), enhances classroom discussion quality, and shifts debriefing practices from instructor-centered toward greater student participation.

0 citationsRead paper

Ensemble of Unsupervised Deep Learning for Clustering Imbalanced Tabular Data

Jul 31, 2026

This study addresses the performance degradation and clustering bias of unsupervised methods on class-imbalanced tabular data. It presents the first systematic evaluation of deep clustering approaches under varying degrees of class imbalance and introduces a novel ensemble mechanism that integrates clustering results from multiple embedding dimensions while combining several high-performing algorithms through majority voting. Evaluated on 16 imbalanced tabular datasets, the proposed method consistently outperforms individual clustering models, achieving notable improvements in Accuracy (ACC), Normalized Mutual Information (NMI), and Adjusted Rand Index (ARI). These gains demonstrate its enhanced capability to recover the underlying true class structure in imbalanced settings.

0 citationsRead paper

Hierarchical Flow Decomposition for Turning Movement Prediction at Signalized Intersections

Apr 10, 2026

This study addresses the challenge of accurately predicting turning movements at signalized intersections, where directional traffic flows exhibit high volatility. To this end, the authors propose the HFD-TM framework, which introduces a novel hierarchical flow decomposition architecture: it first forecasts more stable and interpretable through-movements along traffic corridors and subsequently disaggregates them into individual turning flows. A physics-informed loss function enforcing flow conservation is incorporated to maintain physical consistency. By integrating deep learning with physical constraints, the method achieves a mean absolute error (MAE) of 2.49 vehicles per 15-minute interval on a six-intersection LiDAR dataset in Nashville—outperforming Transformer and GRU baselines by 5.7% and 27.0%, respectively—and trains 12.8 times faster than DCRNN, significantly enhancing both prediction accuracy and computational efficiency.

0 citationsRead paper

Mining Electronic Health Records to Investigate Effectiveness of Ensemble Deep Clustering

Apr 08, 2026

Existing deep clustering methods struggle to effectively identify heart failure patient subtypes from electronic health records (EHR) data. To address this challenge, this work proposes a multidimensional embedding ensemble clustering framework that integrates traditional clustering techniques with deep embeddings. The approach systematically combines autoencoder-derived representations, conventional methods such as K-means, and deep clustering for the first time, while explicitly emphasizing the importance of stratifying models by biological sex. Comprehensive experiments across multiple real-world EHR cohorts and against 14 baseline methods demonstrate that the proposed framework achieves superior overall performance in patient stratification, thereby validating the efficacy and advantage of the ensemble strategy for EHR-driven subtype discovery.

0 citationsRead paper

Physics-Embedded Feature Learning for AI in Medical Imaging

Mar 30, 2026

This study addresses the limited interpretability and robustness of current deep learning models in medical image analysis, which often operate as black boxes and neglect the underlying biophysical mechanisms of tumor growth. To bridge this gap, the authors propose PhysNet, a novel framework that integrates reaction-diffusion equations into intermediate feature layers of a convolutional neural network. Within an end-to-end training paradigm, PhysNet jointly optimizes multi-class tumor classification alongside the inference of latent tumor density fields and associated biophysical parameters—such as diffusion and proliferation rates—ensuring physical consistency without compromising performance. Evaluated on a large-scale brain MRI dataset, PhysNet outperforms MobileNetV2, VGG16, VGG19, and ensemble baselines in both classification accuracy and F1 score, while producing interpretable representations aligned with established medical understanding.

0 citationsRead paper

Recent publications

Latest Papers

Designing PULSE: A Realtime Annotation Tool to Support Simulation Debriefing

Aug 10, 2026

This study addresses the challenges of high instructor workload and low observer engagement in simulation-based education, which often compromise debriefing effectiveness. In collaboration with nursing educators, the authors developed PULSE, a real-time annotation tool that introduces a novel student-led annotation mechanism. PULSE enables observing students to actively document their observations and reflections during simulations, thereby enhancing the structure, depth, and interactivity of debriefings. Grounded in human-centered co-design and informed by the DASH (Debriefing Assessment for Simulation in Healthcare) framework alongside qualitative interviews, the study systematically developed and validated an interactive debriefing support mechanism. Preliminary results demonstrate that PULSE significantly improves DASH scores (p = 0.027, Cohen’s d = 2.05), enhances classroom discussion quality, and shifts debriefing practices from instructor-centered toward greater student participation.

0 citationsRead paper

Ensemble of Unsupervised Deep Learning for Clustering Imbalanced Tabular Data

Jul 31, 2026

This study addresses the performance degradation and clustering bias of unsupervised methods on class-imbalanced tabular data. It presents the first systematic evaluation of deep clustering approaches under varying degrees of class imbalance and introduces a novel ensemble mechanism that integrates clustering results from multiple embedding dimensions while combining several high-performing algorithms through majority voting. Evaluated on 16 imbalanced tabular datasets, the proposed method consistently outperforms individual clustering models, achieving notable improvements in Accuracy (ACC), Normalized Mutual Information (NMI), and Adjusted Rand Index (ARI). These gains demonstrate its enhanced capability to recover the underlying true class structure in imbalanced settings.

0 citationsRead paper

Hierarchical Flow Decomposition for Turning Movement Prediction at Signalized Intersections

Apr 10, 2026

This study addresses the challenge of accurately predicting turning movements at signalized intersections, where directional traffic flows exhibit high volatility. To this end, the authors propose the HFD-TM framework, which introduces a novel hierarchical flow decomposition architecture: it first forecasts more stable and interpretable through-movements along traffic corridors and subsequently disaggregates them into individual turning flows. A physics-informed loss function enforcing flow conservation is incorporated to maintain physical consistency. By integrating deep learning with physical constraints, the method achieves a mean absolute error (MAE) of 2.49 vehicles per 15-minute interval on a six-intersection LiDAR dataset in Nashville—outperforming Transformer and GRU baselines by 5.7% and 27.0%, respectively—and trains 12.8 times faster than DCRNN, significantly enhancing both prediction accuracy and computational efficiency.

0 citationsRead paper

Mining Electronic Health Records to Investigate Effectiveness of Ensemble Deep Clustering

Apr 08, 2026

Existing deep clustering methods struggle to effectively identify heart failure patient subtypes from electronic health records (EHR) data. To address this challenge, this work proposes a multidimensional embedding ensemble clustering framework that integrates traditional clustering techniques with deep embeddings. The approach systematically combines autoencoder-derived representations, conventional methods such as K-means, and deep clustering for the first time, while explicitly emphasizing the importance of stratifying models by biological sex. Comprehensive experiments across multiple real-world EHR cohorts and against 14 baseline methods demonstrate that the proposed framework achieves superior overall performance in patient stratification, thereby validating the efficacy and advantage of the ensemble strategy for EHR-driven subtype discovery.

0 citationsRead paper

Physics-Embedded Feature Learning for AI in Medical Imaging

Mar 30, 2026

This study addresses the limited interpretability and robustness of current deep learning models in medical image analysis, which often operate as black boxes and neglect the underlying biophysical mechanisms of tumor growth. To bridge this gap, the authors propose PhysNet, a novel framework that integrates reaction-diffusion equations into intermediate feature layers of a convolutional neural network. Within an end-to-end training paradigm, PhysNet jointly optimizes multi-class tumor classification alongside the inference of latent tumor density fields and associated biophysical parameters—such as diffusion and proliferation rates—ensuring physical consistency without compromising performance. Evaluated on a large-scale brain MRI dataset, PhysNet outperforms MobileNetV2, VGG16, VGG19, and ensemble baselines in both classification accuracy and F1 score, while producing interpretable representations aligned with established medical understanding.

0 citationsRead paper