Selected work

Representative Papers

A Generic Service-Oriented Function Offloading Framework for Connected Automated Vehicles

May 01, 2025IEEE Robotics and Automation Letters

This work addresses the challenge that connected and autonomous vehicles (CAVs), constrained by limited computational and energy resources, struggle to efficiently execute complex tasks. To this end, the authors propose a service-oriented, generic function offloading framework that integrates multi-access edge computing (MEC) with a location-aware mechanism. The framework dynamically decides whether to process tasks locally or offload them to edge servers, while supporting configurable quality-of-service (QoS) constraints. Its key innovations lie in a location-driven offloading strategy and a service-based architecture, which together enable broad applicability across arbitrary computational tasks and effective scalability in multi-vehicle concurrent scenarios. Experimental results demonstrate that the proposed framework significantly improves computational efficiency—particularly in tasks such as trajectory planning—while consistently meeting specified QoS requirements.

1 citationsRead paper

Span-Level Hallucination Detection for LLM-Generated Answers

Apr 25, 2025

This work addresses factual hallucination in large language model (LLM) text generation by proposing the first English–Arabic bilingual token-level hallucination detection framework. Methodologically, it innovatively integrates semantic role labeling (SRL) with retrieval-augmented textual entailment modeling and introduces a logit-driven token confidence calibration mechanism to enable interpretable span-level hallucination localization. Unlike conventional sentence-level binary classification, this framework achieves finer-grained detection with enhanced precision and attribution capability. Evaluated on the Mu-SHROOM benchmark, it establishes new state-of-the-art performance. Hallucinated spans are rigorously verified via cross-fact-checking using GPT-4 and LLaMA, significantly improving both detection accuracy and interpretability. The framework thus introduces a novel paradigm for multilingual, trustworthy evaluation of LLM outputs.

1 citationsRead paper

Simulation-based Analysis Of Highway Trajectory Planning Using High-Order Polynomial For Highly Automated Driving Function

Oct 07, 20212021 International Conference on Electrical, Computer, Communications and Mechatronics Engineering (ICECCME)

To address insufficient safety and dynamic stability in lane-change trajectory planning for High-Autonomy Driving Functions (HADFs) on highway same-direction dual-lane scenarios, this paper proposes a novel method that deeply integrates high-order polynomial trajectory modeling into a closed-loop behavioral planner. The approach jointly incorporates environmental perception, traffic regulations, and motion constraints to enable context-adaptive, computationally efficient, real-time collision-free trajectory generation. Its key innovation lies in the first-ever construction of a behavior-motion co-designed high-order polynomial interpolation framework—replacing conventional quintic polynomials. MATLAB simulations demonstrate a 32% reduction in peak lateral jerk, a lane-change timing error of less than 0.15 s, and significant improvements in trajectory smoothness and safety.

1 citationsRead paper

How Far from Clinical Deployment? Evaluating the Complete Unsupervised Domain Adaptation Pipeline in Medical Imaging

Aug 12, 2026

This study addresses a critical challenge in the clinical deployment of unsupervised domain adaptation (UDA): the absence of labeled data in the target domain impedes reliable model selection, thereby limiting real-world performance. For the first time, UDA algorithms and unsupervised model selectors are jointly evaluated as an integrated pipeline across 11 cross-domain scenarios spanning 9 medical imaging datasets, involving 10 UDA methods and 13 selection strategies—yielding over 80,000 models analyzed. The findings reveal that model selection constitutes a key bottleneck in UDA adoption: although high-performing adapted models often exist, current unsupervised selectors struggle to identify them consistently. While ensemble learning and minimal target-domain annotations substantially narrow the gap to oracle-level performance, they do not fully eliminate it.

0 citationsRead paper

PaCoNet: Deep Data Extraction for Parallel Coordinates

Aug 06, 2026

This work addresses the longstanding challenge of effectively parsing high-dimensional data from parallel coordinate plots. We propose the first deep learning framework specifically designed for data extraction from parallel coordinates, leveraging a tailored neural network architecture to end-to-end recover line coordinates and reconstruct original data samples directly from images. To support this endeavor, we introduce the first large-scale dataset of parallel coordinate visualizations and demonstrate precise parsing of densely packed, high-dimensional line structures. Experimental results show that our method significantly outperforms general-purpose baselines, achieving, for the first time, accurate and fully automatic reconstruction of individual data samples from parallel coordinate plots. This advance marks a significant step toward deeper integration between visualization and computer vision.

0 citationsRead paper

Recent publications

Latest Papers

How Far from Clinical Deployment? Evaluating the Complete Unsupervised Domain Adaptation Pipeline in Medical Imaging

Aug 12, 2026

This study addresses a critical challenge in the clinical deployment of unsupervised domain adaptation (UDA): the absence of labeled data in the target domain impedes reliable model selection, thereby limiting real-world performance. For the first time, UDA algorithms and unsupervised model selectors are jointly evaluated as an integrated pipeline across 11 cross-domain scenarios spanning 9 medical imaging datasets, involving 10 UDA methods and 13 selection strategies—yielding over 80,000 models analyzed. The findings reveal that model selection constitutes a key bottleneck in UDA adoption: although high-performing adapted models often exist, current unsupervised selectors struggle to identify them consistently. While ensemble learning and minimal target-domain annotations substantially narrow the gap to oracle-level performance, they do not fully eliminate it.

0 citationsRead paper

PaCoNet: Deep Data Extraction for Parallel Coordinates

Aug 06, 2026

This work addresses the longstanding challenge of effectively parsing high-dimensional data from parallel coordinate plots. We propose the first deep learning framework specifically designed for data extraction from parallel coordinates, leveraging a tailored neural network architecture to end-to-end recover line coordinates and reconstruct original data samples directly from images. To support this endeavor, we introduce the first large-scale dataset of parallel coordinate visualizations and demonstrate precise parsing of densely packed, high-dimensional line structures. Experimental results show that our method significantly outperforms general-purpose baselines, achieving, for the first time, accurate and fully automatic reconstruction of individual data samples from parallel coordinate plots. This advance marks a significant step toward deeper integration between visualization and computer vision.

0 citationsRead paper

Bar-JEPA: Extracting Values from Bar Chart with Joint-Embedding Predictive Architecture

Aug 06, 2026

This work addresses the challenging task of automatically recovering underlying numerical values from bar charts, which is hindered by the scarcity and high cost of annotated data. The study introduces, for the first time, a Joint Embedding Predictive Architecture (JEPA) to chart understanding, proposing a self-supervised learning framework that leverages JEPA pretraining to extract semantically rich latent features. These features are then fed into a lightweight decoder to regress tick labels and bar coordinates, thereby reconstructing the original numerical values. The approach requires no large-scale labeled datasets and significantly outperforms end-to-end supervised baselines on numerical extraction tasks, demonstrating both the effectiveness and generalization capability of the learned representations. Code and data are publicly released.

0 citationsRead paper

High-Dimensional Panel Data Models with Interactive Fixed Effects: Beyond the Linear Case

Aug 03, 2026

This study addresses the challenge of nonlinear modeling in high-dimensional panel data, where the number of covariates may exceed the sample size and interactive fixed effects are present. The authors propose an additive nonparametric model that allows each covariate to influence the response through an unknown nonlinear function. Methodologically, they extend existing high-dimensional linear panel models to this nonlinear additive setting by integrating regularization techniques with high-dimensional asymptotic theory, establishing a unified estimation framework applicable to both small-T and large-T scenarios for the first time. Theoretical analysis provides convergence rates of the proposed estimator under both settings, while Monte Carlo simulations confirm its finite-sample performance. An empirical application further demonstrates the practical utility of the approach.

0 citationsRead paper

Towards Practical Algorithm Selection for Unsupervised Domain Adaptation in Medical Imaging

Jul 30, 2026

This work addresses the challenge of model selection in unsupervised domain adaptation for medical imaging, where the absence of labels in the target domain hinders effective algorithm and hyperparameter tuning. The authors propose a novel, label-free joint model selection criterion that aggregates candidate models across diverse algorithms and hyperparameter configurations by leveraging multiple unlabeled signals. This approach constructs a two-stage consistency-based reference prediction and selects the model whose predictions align most closely with this reference for deployment. To the best of our knowledge, this is the first method to enable unified selection of both algorithms and hyperparameters without requiring target-domain labels. Evaluated across eight medical imaging datasets and seven clinical transfer scenarios, the proposed method consistently outperforms existing approaches, demonstrating strong generalization and stability.

0 citationsRead paper