Selected work

Representative Papers

Kyrtos: A methodology for automatic deep analysis of graphic charts with curves in technical documents

Jan 01, 2025Pattern Recognition

This work addresses the challenge of automatically achieving deep semantic understanding of curve-based charts in technical documentation, which often lack structured semantic representations. To this end, the authors propose Kyrtos, a novel method that first segments curves by clustering inflection points and analyzing behavioral features such as direction and trend. It then constructs an attributed graph and generates corresponding natural language descriptions, ultimately mapping the representation end-to-end into a stochastic Petri net (SPN) to capture the chart’s internal functional logic. This study presents the first approach to jointly represent curve structures through attributed graphs and natural language while enabling direct conversion to SPNs. Experimental results demonstrate that Kyrtos accurately reconstructs both structural and semantic relationships in multi-function curve charts, significantly advancing the deep semantic understanding of technical diagrams.

1 citationsRead paper

Blockchain-Driven AI-Enhanced Post-Quantum Multivariate Identity-based Signature and Privacy-Preserving Data Aggregation Scheme for Fog-enabled Flying Ad-Hoc Networks

Apr 20, 2026

This work addresses the challenges of security, privacy, and resource constraints in fog computing–enabled flying ad hoc networks (FANETs) under quantum threats by proposing a novel key management framework that integrates blockchain and artificial intelligence. The framework uniquely combines post-quantum multivariate identity-based signatures, zero-knowledge proofs, and a hierarchical blockchain consensus mechanism to enable efficient and secure key establishment, privacy-preserving data aggregation, and integrity verification. It further leverages a fog–cloud协同 architecture to support intelligent data analytics. NS-3 simulations demonstrate that, compared to existing approaches, the proposed method significantly reduces communication overhead and enhances the efficiency of data aggregation and verification, offering distinct advantages in computational cost, quantum resistance, and scalability.

0 citationsRead paper

K-Means Based TinyML Anomaly Detection and Distributed Model Reuse via the Distributed Internet of Learning (DIoL)

Mar 28, 2026

This work addresses anomaly detection on resource-constrained microcontrollers by proposing a “Train Once, Share Everywhere” (TOSE) paradigm that eliminates the need for per-device retraining. The approach leverages a lightweight K-Means clustering algorithm to perform local feature extraction and adaptive threshold estimation, integrated within a Distributed Internet-of-Learning (DIoL) framework. By encoding models into a textual representation, the method enables direct cross-device model reuse without additional training. Experimental validation on a dual-device prototype demonstrates that the shared model maintains consistent detection performance, achieves inference speeds comparable to independently deployed models, and incurs negligible parsing overhead. This significantly enhances deployment efficiency and system scalability in edge environments with limited computational resources.

0 citationsRead paper

Handling Extreme Class Imbalance: Using GANs in Data Augmentation for Suicide Prediction

Oct 20, 2025

To address extreme class imbalance caused by severe scarcity of positive samples in suicide prediction, this paper proposes a generative adversarial network (GAN)-based data augmentation method to synthesize high-fidelity positive instances and mitigate data sparsity. Unlike conventional oversampling techniques, GAN-based augmentation better preserves the underlying high-dimensional feature distribution. Evaluated on real-world clinical text data, the approach is integrated with logistic regression, random forest (RF), and support vector machine (SVM) classifiers. Results show that RF achieves 0.98 weighted precision, 0.99 weighted recall, and 0.99 weighted F1-score after GAN augmentation—significantly outperforming all baselines. Ablation studies confirm that the synthetic samples critically enhance generalization under limited positive-label conditions. This work establishes a reproducible and scalable generative data augmentation paradigm for predicting rare, high-risk events.

0 citationsRead paper

Improving Knowledge Graph Understanding with Contextual Views

Aug 04, 2025

To address insufficient utilization of ontology information in knowledge graph (KG) navigation and exploration, this paper proposes an ontology-based multimodal interactive exploration framework. Methodologically, we design a context-aware, ontology-driven view mechanism that integrates the schema layer, instance layer (types and neighborhoods), and geospatial dimensions, implemented via a modular frontend architecture incorporating ontology parsing, dynamic neighborhood extraction, geospatial mapping, and scalable visualization. Our key contribution is the first deep integration of ontological semantics into a multi-granularity interactive pipeline, enabling consistent browsing across schema, instance, and spatial layers. User evaluation demonstrates that the framework significantly reduces navigational cognitive load (p < 0.01), improves target entity discovery efficiency by 37%, and supports real-time interaction over KGs containing up to ten million triples, with demonstrated cross-domain deployability.

0 citationsRead paper

Recent publications

Latest Papers

Blockchain-Driven AI-Enhanced Post-Quantum Multivariate Identity-based Signature and Privacy-Preserving Data Aggregation Scheme for Fog-enabled Flying Ad-Hoc Networks

Apr 20, 2026

This work addresses the challenges of security, privacy, and resource constraints in fog computing–enabled flying ad hoc networks (FANETs) under quantum threats by proposing a novel key management framework that integrates blockchain and artificial intelligence. The framework uniquely combines post-quantum multivariate identity-based signatures, zero-knowledge proofs, and a hierarchical blockchain consensus mechanism to enable efficient and secure key establishment, privacy-preserving data aggregation, and integrity verification. It further leverages a fog–cloud协同 architecture to support intelligent data analytics. NS-3 simulations demonstrate that, compared to existing approaches, the proposed method significantly reduces communication overhead and enhances the efficiency of data aggregation and verification, offering distinct advantages in computational cost, quantum resistance, and scalability.

0 citationsRead paper

K-Means Based TinyML Anomaly Detection and Distributed Model Reuse via the Distributed Internet of Learning (DIoL)

Mar 28, 2026

This work addresses anomaly detection on resource-constrained microcontrollers by proposing a “Train Once, Share Everywhere” (TOSE) paradigm that eliminates the need for per-device retraining. The approach leverages a lightweight K-Means clustering algorithm to perform local feature extraction and adaptive threshold estimation, integrated within a Distributed Internet-of-Learning (DIoL) framework. By encoding models into a textual representation, the method enables direct cross-device model reuse without additional training. Experimental validation on a dual-device prototype demonstrates that the shared model maintains consistent detection performance, achieves inference speeds comparable to independently deployed models, and incurs negligible parsing overhead. This significantly enhances deployment efficiency and system scalability in edge environments with limited computational resources.

0 citationsRead paper

Handling Extreme Class Imbalance: Using GANs in Data Augmentation for Suicide Prediction

Oct 20, 2025

To address extreme class imbalance caused by severe scarcity of positive samples in suicide prediction, this paper proposes a generative adversarial network (GAN)-based data augmentation method to synthesize high-fidelity positive instances and mitigate data sparsity. Unlike conventional oversampling techniques, GAN-based augmentation better preserves the underlying high-dimensional feature distribution. Evaluated on real-world clinical text data, the approach is integrated with logistic regression, random forest (RF), and support vector machine (SVM) classifiers. Results show that RF achieves 0.98 weighted precision, 0.99 weighted recall, and 0.99 weighted F1-score after GAN augmentation—significantly outperforming all baselines. Ablation studies confirm that the synthetic samples critically enhance generalization under limited positive-label conditions. This work establishes a reproducible and scalable generative data augmentation paradigm for predicting rare, high-risk events.

0 citationsRead paper

Improving Knowledge Graph Understanding with Contextual Views

Aug 04, 2025

To address insufficient utilization of ontology information in knowledge graph (KG) navigation and exploration, this paper proposes an ontology-based multimodal interactive exploration framework. Methodologically, we design a context-aware, ontology-driven view mechanism that integrates the schema layer, instance layer (types and neighborhoods), and geospatial dimensions, implemented via a modular frontend architecture incorporating ontology parsing, dynamic neighborhood extraction, geospatial mapping, and scalable visualization. Our key contribution is the first deep integration of ontological semantics into a multi-granularity interactive pipeline, enabling consistent browsing across schema, instance, and spatial layers. User evaluation demonstrates that the framework significantly reduces navigational cognitive load (p < 0.01), improves target entity discovery efficiency by 37%, and supports real-time interaction over KGs containing up to ten million triples, with demonstrated cross-domain deployability.

0 citationsRead paper

Enhancing IoT Intrusion Detection Systems through Adversarial Training

Jul 25, 2025

To address widespread security vulnerabilities in Internet of Things (IoT) devices, this paper proposes a lightweight intrusion detection system (IDS) based on adversarial training. Methodologically, it introduces the Fast Gradient Sign Method (FGSM) into IoT intrusion detection for the first time, integrating high-dimensional network flow features from the NF-ToN-IoT v2 dataset within a distributed preprocessing and adversarial sample generation framework, and employs XGBoost as the base classifier for adversarial training. The key contribution lies in adapting the FGSM mechanism to IoT traffic characteristics, enabling effective modeling and robust detection of complex, stealthy attacks. Experimental results demonstrate that the model achieves 95.3% accuracy on clean test data and maintains 94.5% accuracy under adversarial perturbations—significantly outperforming baseline methods—thereby validating its superior robustness and practical applicability for real-world IoT security.

0 citationsRead paper