Selected work

Representative Papers

"Blockchain-Enabled Zero Trust Framework for Securing FinTech Ecosystems Against Insider Threats and Cyber Attacks"

Jul 26, 2025

Financial technology (FinTech) systems face escalating internal threats and advanced persistent threats (APTs), rendering traditional perimeter-based security models ineffective. To address this, this paper proposes a blockchain-enabled zero-trust security framework grounded in the principle of “never trust, always verify,” enabling dynamic access control and micro-segmentation. The framework innovatively leverages blockchain as a unified policy engine, enforcement point, and tamper-proof storage layer—thereby eliminating single points of failure. It integrates Ethereum smart contracts, multi-factor authentication (MFA), role-based access control (RBAC), and just-in-time (JIT) privilege management. Security validation of the decentralized application (DApp) is rigorously conducted using STRIDE threat modeling. Experimental evaluation on a 200-node network demonstrates significant security enhancement, bounded latency overhead, and native support for Layer-2 scalability optimizations.

0 citationsRead paper

CLAIM: An Intent-Driven Multi-Agent Framework for Analyzing Manipulation in Courtroom Dialogues

Jun 04, 2025

Manipulative behavior detection in courtroom dialogues has long suffered from a lack of systematic investigation and high-quality, annotated data. To address this gap, we propose CLAIM—a novel, intent-driven, two-stage multi-agent framework for manipulative behavior detection and attribution in judicial settings. Our method integrates dialogue state tracking, legal-domain fine-tuned language models, fine-grained sequence labeling, and relation extraction. We introduce LegalCon, the first large-scale, long-context, expert-annotated courtroom dialogue dataset (1,063 trials), with explicit annotations of manipulators, manipulation tactics, and context-dependent relational dependencies. On LegalCon, CLAIM achieves 82.3% F1 for manipulation detection and 79.6% accuracy for manipulator identification—significantly outperforming single-model baselines. We publicly release both code and data, establishing the first benchmark for modeling manipulative discourse in legal NLP and advancing explainable, justice-oriented AI systems.

0 citationsRead paper

MECASA: Motor Execution Classification using Additive Self-Attention for Hybrid EEG-fNIRS Data

Jan 09, 2025

This study addresses the challenge of accurate discrimination between motor execution states (rest vs. task), overcoming the limitations of low signal-to-noise ratio in EEG and limited spatiotemporal resolution in fNIRS when used individually. We propose an end-to-end EEG–fNIRS bimodal collaborative modeling framework, whose core innovations are a Convolutional Additive Self-Attention (CASA) module and a dual-stream hybrid feature fusion mechanism, incorporating 1D convolution, additive self-attention, and OD128 upsampled input support. Built upon the CAS-ViT architecture, our method achieves significantly higher fusion accuracy than unimodal baselines on the SMR Hybrid BCI dataset. Notably, fNIRS alone outperforms EEG, and optimal EEG embedding dimensions lie within 64–128. The proposed framework establishes a robust, interpretable multimodal paradigm for motor brain–computer interfaces and underlying neurocognitive mechanism investigation.

0 citationsRead paper

Recent publications

Latest Papers

"Blockchain-Enabled Zero Trust Framework for Securing FinTech Ecosystems Against Insider Threats and Cyber Attacks"

Jul 26, 2025

Financial technology (FinTech) systems face escalating internal threats and advanced persistent threats (APTs), rendering traditional perimeter-based security models ineffective. To address this, this paper proposes a blockchain-enabled zero-trust security framework grounded in the principle of “never trust, always verify,” enabling dynamic access control and micro-segmentation. The framework innovatively leverages blockchain as a unified policy engine, enforcement point, and tamper-proof storage layer—thereby eliminating single points of failure. It integrates Ethereum smart contracts, multi-factor authentication (MFA), role-based access control (RBAC), and just-in-time (JIT) privilege management. Security validation of the decentralized application (DApp) is rigorously conducted using STRIDE threat modeling. Experimental evaluation on a 200-node network demonstrates significant security enhancement, bounded latency overhead, and native support for Layer-2 scalability optimizations.

0 citationsRead paper

CLAIM: An Intent-Driven Multi-Agent Framework for Analyzing Manipulation in Courtroom Dialogues

Jun 04, 2025

Manipulative behavior detection in courtroom dialogues has long suffered from a lack of systematic investigation and high-quality, annotated data. To address this gap, we propose CLAIM—a novel, intent-driven, two-stage multi-agent framework for manipulative behavior detection and attribution in judicial settings. Our method integrates dialogue state tracking, legal-domain fine-tuned language models, fine-grained sequence labeling, and relation extraction. We introduce LegalCon, the first large-scale, long-context, expert-annotated courtroom dialogue dataset (1,063 trials), with explicit annotations of manipulators, manipulation tactics, and context-dependent relational dependencies. On LegalCon, CLAIM achieves 82.3% F1 for manipulation detection and 79.6% accuracy for manipulator identification—significantly outperforming single-model baselines. We publicly release both code and data, establishing the first benchmark for modeling manipulative discourse in legal NLP and advancing explainable, justice-oriented AI systems.

0 citationsRead paper

MECASA: Motor Execution Classification using Additive Self-Attention for Hybrid EEG-fNIRS Data

Jan 09, 2025

This study addresses the challenge of accurate discrimination between motor execution states (rest vs. task), overcoming the limitations of low signal-to-noise ratio in EEG and limited spatiotemporal resolution in fNIRS when used individually. We propose an end-to-end EEG–fNIRS bimodal collaborative modeling framework, whose core innovations are a Convolutional Additive Self-Attention (CASA) module and a dual-stream hybrid feature fusion mechanism, incorporating 1D convolution, additive self-attention, and OD128 upsampled input support. Built upon the CAS-ViT architecture, our method achieves significantly higher fusion accuracy than unimodal baselines on the SMR Hybrid BCI dataset. Notably, fNIRS alone outperforms EEG, and optimal EEG embedding dimensions lie within 64–128. The proposed framework establishes a robust, interpretable multimodal paradigm for motor brain–computer interfaces and underlying neurocognitive mechanism investigation.

0 citationsRead paper