Selected work

Representative Papers

A Stochastic Optimization Framework for RIS-Aided Wireless Network Design

Jul 30, 2026

This work addresses the challenge of rapidly escalating computational complexity in large-scale non-convex reconfigurable intelligent surface (RIS) configuration optimization, which intensifies with the number of scattering elements and architectural intricacy. To tackle this, the paper proposes a stochastic optimization framework that integrates continuous cross-entropy (CE) methods with Metropolis–Hastings (MH) sampling. The approach operates directly on continuous variables and incorporates relaxation and projection mechanisms to accommodate discrete RIS configurations, making it applicable to both nearly passive and active RIS architectures for optimizing spectral efficiency and energy efficiency. As the first systematic application of continuous stochastic optimization to RIS network design, the proposed method transcends the limitations of conventional discrete optimization, offering theoretical guarantees on convergence and computational efficiency. In representative scenarios, it achieves performance comparable to or better than state-of-the-art deterministic algorithms while reducing runtime by up to an order of magnitude.

0 citationsRead paper

Uncertainty-Aware Mapping from 3D Keypoints to Anatomical Landmarks for Markerless Biomechanics

Mar 27, 2026

This study addresses a critical limitation in existing markerless biomechanical methods: the neglect of estimation uncertainty when mapping 3D keypoints to anatomical landmarks, resulting in a lack of per-frame quality control. The work proposes the first temporal learning framework that explicitly models uncertainties arising from both observational noise and model limitations, leveraging them as an intrinsic mechanism for automatic quality assessment to quantify mapping confidence. Through comprehensive analyses—including error–uncertainty correlation, risk–coverage evaluation, and anomaly detection—the authors demonstrate that model uncertainty is a dominant indicator of mapping failure. Experimental results show a strong correlation between predicted uncertainty and landmark error (Spearman ρ ≈ 0.63); at 10% coverage, the average error drops to 16.8 mm, and the method achieves a ROC-AUC of 0.92 in detecting severe errors exceeding 50 mm.

0 citationsRead paper

Robust measures of dispersion for circular data with an anomaly detection rule

Mar 01, 2026

This study addresses the lack of robustness in estimating dispersion for circular data under outlier contamination. For the first time, it extends three linear robust dispersion measures to the circular domain, analyzing their robustness through influence functions and relative deviation curves. The authors develop high-breakdown-point, high-efficiency parameter estimators tailored for von Mises and wrapped normal distributions. Furthermore, they propose a novel circular anomaly detection method and introduce circular violin plots for intuitive outlier visualization. Extensive Monte Carlo simulations and experiments on three real-world datasets demonstrate that the proposed approach significantly outperforms existing methods in both estimation accuracy and outlier detection capability.

0 citationsRead paper

Handling Missing Modalities in Multimodal Survival Prediction for Non-Small Cell Lung Cancer

Jan 15, 2026

This study addresses the challenge of multimodal survival prediction in non-small cell lung cancer, where missing data across CT imaging, whole-slide pathology images (WSI), and clinical records often hinder the clinical deployment of deep learning models. To overcome this limitation, the authors propose a missingness-aware multimodal survival prediction framework that leverages foundation models to extract modality-specific features and introduces a missingness-aware encoding mechanism. This design enables the model to adaptively utilize available information without discarding incomplete samples or resorting to imputation, while dynamically adjusting each modality’s contribution during intermediate fusion. Experimental results demonstrate that the proposed approach significantly outperforms both unimodal baselines and early/late fusion strategies under naturally occurring missing modalities, achieving a C-index of 73.30 when fusing WSI and clinical data, thereby confirming its effectiveness and robustness.

0 citationsRead paper

Character Detection using YOLO for Writer Identification in multiple Medieval books

Sep 08, 20252025 IEEE International Conference on Cyber Humanities (IEEE-CH)

This study addresses the challenge of identifying scribes in medieval manuscripts by proposing a YOLOv5-based object detection approach to localize and extract discriminative characters—such as the letter “a”—for supporting paleographic dating and stylistic evolution analysis. Departing from conventional template-matching and CNN-based classification pipelines, this work is the first to adapt YOLOv5 to the task of historical handwriting identification. The method substantially increases both the quantity and reliability of detected characters, while a confidence-thresholding mechanism enables robust recognition and rejection of unseen manuscripts. This enhances the system’s generalization capability and improves the accuracy of subsequent writer classification.

0 citationsRead paper

Recent publications

Latest Papers

A Stochastic Optimization Framework for RIS-Aided Wireless Network Design

Jul 30, 2026

This work addresses the challenge of rapidly escalating computational complexity in large-scale non-convex reconfigurable intelligent surface (RIS) configuration optimization, which intensifies with the number of scattering elements and architectural intricacy. To tackle this, the paper proposes a stochastic optimization framework that integrates continuous cross-entropy (CE) methods with Metropolis–Hastings (MH) sampling. The approach operates directly on continuous variables and incorporates relaxation and projection mechanisms to accommodate discrete RIS configurations, making it applicable to both nearly passive and active RIS architectures for optimizing spectral efficiency and energy efficiency. As the first systematic application of continuous stochastic optimization to RIS network design, the proposed method transcends the limitations of conventional discrete optimization, offering theoretical guarantees on convergence and computational efficiency. In representative scenarios, it achieves performance comparable to or better than state-of-the-art deterministic algorithms while reducing runtime by up to an order of magnitude.

0 citationsRead paper

Uncertainty-Aware Mapping from 3D Keypoints to Anatomical Landmarks for Markerless Biomechanics

Mar 27, 2026

This study addresses a critical limitation in existing markerless biomechanical methods: the neglect of estimation uncertainty when mapping 3D keypoints to anatomical landmarks, resulting in a lack of per-frame quality control. The work proposes the first temporal learning framework that explicitly models uncertainties arising from both observational noise and model limitations, leveraging them as an intrinsic mechanism for automatic quality assessment to quantify mapping confidence. Through comprehensive analyses—including error–uncertainty correlation, risk–coverage evaluation, and anomaly detection—the authors demonstrate that model uncertainty is a dominant indicator of mapping failure. Experimental results show a strong correlation between predicted uncertainty and landmark error (Spearman ρ ≈ 0.63); at 10% coverage, the average error drops to 16.8 mm, and the method achieves a ROC-AUC of 0.92 in detecting severe errors exceeding 50 mm.

0 citationsRead paper

Robust measures of dispersion for circular data with an anomaly detection rule

Mar 01, 2026

This study addresses the lack of robustness in estimating dispersion for circular data under outlier contamination. For the first time, it extends three linear robust dispersion measures to the circular domain, analyzing their robustness through influence functions and relative deviation curves. The authors develop high-breakdown-point, high-efficiency parameter estimators tailored for von Mises and wrapped normal distributions. Furthermore, they propose a novel circular anomaly detection method and introduce circular violin plots for intuitive outlier visualization. Extensive Monte Carlo simulations and experiments on three real-world datasets demonstrate that the proposed approach significantly outperforms existing methods in both estimation accuracy and outlier detection capability.

0 citationsRead paper

Handling Missing Modalities in Multimodal Survival Prediction for Non-Small Cell Lung Cancer

Jan 15, 2026

This study addresses the challenge of multimodal survival prediction in non-small cell lung cancer, where missing data across CT imaging, whole-slide pathology images (WSI), and clinical records often hinder the clinical deployment of deep learning models. To overcome this limitation, the authors propose a missingness-aware multimodal survival prediction framework that leverages foundation models to extract modality-specific features and introduces a missingness-aware encoding mechanism. This design enables the model to adaptively utilize available information without discarding incomplete samples or resorting to imputation, while dynamically adjusting each modality’s contribution during intermediate fusion. Experimental results demonstrate that the proposed approach significantly outperforms both unimodal baselines and early/late fusion strategies under naturally occurring missing modalities, achieving a C-index of 73.30 when fusing WSI and clinical data, thereby confirming its effectiveness and robustness.

0 citationsRead paper

Character Detection using YOLO for Writer Identification in multiple Medieval books

Sep 08, 20252025 IEEE International Conference on Cyber Humanities (IEEE-CH)

This study addresses the challenge of identifying scribes in medieval manuscripts by proposing a YOLOv5-based object detection approach to localize and extract discriminative characters—such as the letter “a”—for supporting paleographic dating and stylistic evolution analysis. Departing from conventional template-matching and CNN-based classification pipelines, this work is the first to adapt YOLOv5 to the task of historical handwriting identification. The method substantially increases both the quantity and reliability of detected characters, while a confidence-thresholding mechanism enables robust recognition and rejection of unseen manuscripts. This enhances the system’s generalization capability and improves the accuracy of subsequent writer classification.

0 citationsRead paper