Selected work

Representative Papers

Discrete-Time Survival Analysis for Heart Failure Mortality Prediction

Aug 04, 2026

This study addresses critical methodological flaws in existing heart failure mortality prediction models, particularly target leakage arising from the inappropriate inclusion of follow-up duration as a predictor and the neglect of right-censoring structures, both of which induce substantial evaluation bias. By adopting a discrete-time person-period framework that transforms clinical records into interval-level binary events, the authors systematically compare Cox regression, complementary log-log generalized linear models (GLMs), generalized additive models (GAMs), random forests, XGBoost, random survival forests, and DeepSurv. Results demonstrate that person-period GLMs accurately reproduce the hazard ratios and concordance indices of Cox models, while GAMs achieve the best trade-off between modeling nonlinear effects and generalization performance, yielding superior predictive accuracy. Crucially, erroneously incorporating follow-up time inflates the AUC from 0.73 to nearly 1.00, clearly exposing and quantifying the severity of target leakage.

0 citationsRead paper

MQTT Across a Raspberry Pi 5 IoT Network Utilizing Quantum-resistant Signature Algorithms

May 13, 2026

This study addresses the vulnerability of traditional MQTT communication—secured via TLS—to emerging quantum computing threats, which jeopardizes the long-term security of Internet of Things (IoT) systems. For the first time, the authors integrate FALCON, a lattice-based post-quantum digital signature algorithm selected by NIST, into a real-world MQTT network deployed on Raspberry Pi 5 devices to enable secure authentication between clients and brokers. Through empirical measurements, they systematically evaluate the latency and computational overhead associated with FALCON’s signing and verification operations, thereby assessing its feasibility and performance trade-offs on resource-constrained hardware. The results demonstrate that post-quantum cryptography can be practically deployed in lightweight IoT environments without prohibitive resource demands, offering a viable path toward quantum-resistant IoT security.

0 citationsRead paper

Unsupervised Multi-agent and Single-agent Perception from Cooperative Views

Apr 06, 2026

This work addresses the challenge that existing unsupervised methods struggle to simultaneously handle both multi-agent and single-agent 3D perception tasks. It proposes the first unified unsupervised framework that enhances point cloud density by sharing LiDAR data among agents and leverages cooperative viewpoints to generate high-quality pseudo-labels, enabling joint optimization of single-view and multi-view 3D object detection and classification. The method introduces a learned proposal refinement filter, a progressive proposal stabilization module, and a cross-view consistency learning mechanism. Evaluated on the V2V4Real and OPV2V datasets under fully unsupervised settings, the proposed approach significantly outperforms current state-of-the-art methods and achieves the best performance across both perception tasks.

0 citationsRead paper

DarkDriving: A Real-World Day and Night Aligned Dataset for Autonomous Driving in the Dark Environment

Mar 17, 2026

This work addresses the significant degradation of visual perception performance in autonomous driving under low-light conditions, a challenge exacerbated by the lack of accurately aligned day–night image pairs from real-world dynamic driving scenarios. To bridge this gap, the authors propose an automated data collection framework based on Trajectory Tracking and Pose Matching (TTPM), enabling the creation of DarkDriving—the first real-world day–night image dataset with centimeter-level alignment, captured in a 69-acre closed test track. DarkDriving comprises 9,538 high-precision image pairs accompanied by human-annotated 2D bounding boxes. The dataset facilitates multi-task evaluation of low-light enhancement and its downstream impact on 2D/3D object detection, while also introducing four novel low-light enhancement tasks tailored to autonomous driving perception. Experiments demonstrate that models trained on DarkDriving not only achieve superior low-light enhancement but also generalize effectively to improve perception in other low-light driving benchmarks such as nuScenes.

0 citationsRead paper

Agentic AI in Healthcare and Medicine: A Seven-Dimensional Taxonomy for Empirical Evaluation of LLM-Based Agents

Feb 04, 2026IEEE Access

This study addresses the absence of a systematic evaluation framework for large language model (LLM) agents in healthcare. The authors propose the first seven-dimensional assessment framework tailored to medical AI agents, encompassing cognition, knowledge management, interaction, adaptive learning, safety and ethics, agent architecture, and core clinical tasks. This framework is operationalized into 29 measurable sub-dimensions and applied through a systematic literature review of 49 studies, using a three-tier annotation scheme (fully/partially/not implemented) for quantitative mapping and co-occurrence analysis. Findings reveal that external knowledge integration is widely implemented (76% fully), whereas event-triggered activation (92% not implemented) and drift detection (98% not implemented) are critically underdeveloped. Multi-agent architectures dominate (82% fully), yet action-oriented tasks such as treatment planning remain notably underexplored.

0 citationsRead paper

Recent publications

Latest Papers

Discrete-Time Survival Analysis for Heart Failure Mortality Prediction

Aug 04, 2026

This study addresses critical methodological flaws in existing heart failure mortality prediction models, particularly target leakage arising from the inappropriate inclusion of follow-up duration as a predictor and the neglect of right-censoring structures, both of which induce substantial evaluation bias. By adopting a discrete-time person-period framework that transforms clinical records into interval-level binary events, the authors systematically compare Cox regression, complementary log-log generalized linear models (GLMs), generalized additive models (GAMs), random forests, XGBoost, random survival forests, and DeepSurv. Results demonstrate that person-period GLMs accurately reproduce the hazard ratios and concordance indices of Cox models, while GAMs achieve the best trade-off between modeling nonlinear effects and generalization performance, yielding superior predictive accuracy. Crucially, erroneously incorporating follow-up time inflates the AUC from 0.73 to nearly 1.00, clearly exposing and quantifying the severity of target leakage.

0 citationsRead paper

MQTT Across a Raspberry Pi 5 IoT Network Utilizing Quantum-resistant Signature Algorithms

May 13, 2026

This study addresses the vulnerability of traditional MQTT communication—secured via TLS—to emerging quantum computing threats, which jeopardizes the long-term security of Internet of Things (IoT) systems. For the first time, the authors integrate FALCON, a lattice-based post-quantum digital signature algorithm selected by NIST, into a real-world MQTT network deployed on Raspberry Pi 5 devices to enable secure authentication between clients and brokers. Through empirical measurements, they systematically evaluate the latency and computational overhead associated with FALCON’s signing and verification operations, thereby assessing its feasibility and performance trade-offs on resource-constrained hardware. The results demonstrate that post-quantum cryptography can be practically deployed in lightweight IoT environments without prohibitive resource demands, offering a viable path toward quantum-resistant IoT security.

0 citationsRead paper

Unsupervised Multi-agent and Single-agent Perception from Cooperative Views

Apr 06, 2026

This work addresses the challenge that existing unsupervised methods struggle to simultaneously handle both multi-agent and single-agent 3D perception tasks. It proposes the first unified unsupervised framework that enhances point cloud density by sharing LiDAR data among agents and leverages cooperative viewpoints to generate high-quality pseudo-labels, enabling joint optimization of single-view and multi-view 3D object detection and classification. The method introduces a learned proposal refinement filter, a progressive proposal stabilization module, and a cross-view consistency learning mechanism. Evaluated on the V2V4Real and OPV2V datasets under fully unsupervised settings, the proposed approach significantly outperforms current state-of-the-art methods and achieves the best performance across both perception tasks.

0 citationsRead paper

DarkDriving: A Real-World Day and Night Aligned Dataset for Autonomous Driving in the Dark Environment

Mar 17, 2026

This work addresses the significant degradation of visual perception performance in autonomous driving under low-light conditions, a challenge exacerbated by the lack of accurately aligned day–night image pairs from real-world dynamic driving scenarios. To bridge this gap, the authors propose an automated data collection framework based on Trajectory Tracking and Pose Matching (TTPM), enabling the creation of DarkDriving—the first real-world day–night image dataset with centimeter-level alignment, captured in a 69-acre closed test track. DarkDriving comprises 9,538 high-precision image pairs accompanied by human-annotated 2D bounding boxes. The dataset facilitates multi-task evaluation of low-light enhancement and its downstream impact on 2D/3D object detection, while also introducing four novel low-light enhancement tasks tailored to autonomous driving perception. Experiments demonstrate that models trained on DarkDriving not only achieve superior low-light enhancement but also generalize effectively to improve perception in other low-light driving benchmarks such as nuScenes.

0 citationsRead paper

Agentic AI in Healthcare and Medicine: A Seven-Dimensional Taxonomy for Empirical Evaluation of LLM-Based Agents

Feb 04, 2026IEEE Access

This study addresses the absence of a systematic evaluation framework for large language model (LLM) agents in healthcare. The authors propose the first seven-dimensional assessment framework tailored to medical AI agents, encompassing cognition, knowledge management, interaction, adaptive learning, safety and ethics, agent architecture, and core clinical tasks. This framework is operationalized into 29 measurable sub-dimensions and applied through a systematic literature review of 49 studies, using a three-tier annotation scheme (fully/partially/not implemented) for quantitative mapping and co-occurrence analysis. Findings reveal that external knowledge integration is widely implemented (76% fully), whereas event-triggered activation (92% not implemented) and drift detection (98% not implemented) are critically underdeveloped. Multi-agent architectures dominate (82% fully), yet action-oriented tasks such as treatment planning remain notably underexplored.

0 citationsRead paper