Selected work

Representative Papers

TMR-GGNN: Credit Card Fraud Detection based on Time-Aware Multi-Relational Guided Graph Neural Network

Jun 16, 2026

This work addresses the challenges of extreme class imbalance, dynamically evolving fraud patterns, and complex transactional relationships in credit card fraud detection by proposing a temporal-aware multi-relational graph neural network. The method constructs a dynamic multi-relational graph and employs a temporal relational attention mechanism to adaptively model both semantic and temporal dependencies among transactions. To enhance discrimination of rare fraud patterns, contrastive learning is integrated into the decoder. The model is jointly optimized using InfoNCE loss and Focal Loss, effectively mitigating the class imbalance issue. Experimental results demonstrate that the proposed approach significantly improves fraud detection accuracy, reduces false negatives, and exhibits superior generalization performance under highly imbalanced conditions.

0 citationsRead paper

Explainable post-training bias mitigation with distribution-based fairness metrics

Apr 01, 2025

This paper addresses the challenge of post-training fairness adjustment without model retraining. Methodologically, it introduces a novel distribution-driven differentiable fairness constraint optimization framework, designs an interpretable global bias metric family compatible with diverse models (e.g., gradient-boosted trees), and establishes a continuous, controllable post-processing paradigm for fairness calibration. Technically, it integrates distribution alignment constraints, interpretability regularization, and bias calibration. Evaluated on multiple benchmark datasets, the approach maintains high predictive accuracy while improving demographic parity (DP) and equalized odds (EO) by 30–50% over state-of-the-art post-processing methods. Moreover, it enables global bias attribution with human-interpretable explanations, offering both quantitative fairness enhancement and qualitative insight into bias sources.

0 citationsRead paper

Agentic AI Systems Applied to tasks in Financial Services: Modeling and model risk management crews

Feb 08, 2025

To address the challenge of poor integration between model development and model risk management (MRM) in financial services, this paper proposes the first multi-agent collaborative framework jointly optimized for financial modeling and MRM. Methodologically, it orchestrates two specialized agent teams—leveraging large language models (LLMs) for task decomposition, domain-knowledge-enhanced prompting, structured documentation generation, and automated model validation—to enable end-to-end autonomous collaboration across data exploration, feature engineering, model training, regulatory compliance review, reproducibility verification, and conceptual soundness assessment. The key contribution lies in the first deep coupling of modeling and MRM within a unified multi-agent architecture, ensuring both model performance and stringent regulatory adherence. Experiments on credit card fraud detection, credit approval, and portfolio credit risk modeling demonstrate significant improvements in modeling efficiency, interpretability, and regulatory compliance.

0 citationsRead paper

Recent publications

Latest Papers

TMR-GGNN: Credit Card Fraud Detection based on Time-Aware Multi-Relational Guided Graph Neural Network

Jun 16, 2026

This work addresses the challenges of extreme class imbalance, dynamically evolving fraud patterns, and complex transactional relationships in credit card fraud detection by proposing a temporal-aware multi-relational graph neural network. The method constructs a dynamic multi-relational graph and employs a temporal relational attention mechanism to adaptively model both semantic and temporal dependencies among transactions. To enhance discrimination of rare fraud patterns, contrastive learning is integrated into the decoder. The model is jointly optimized using InfoNCE loss and Focal Loss, effectively mitigating the class imbalance issue. Experimental results demonstrate that the proposed approach significantly improves fraud detection accuracy, reduces false negatives, and exhibits superior generalization performance under highly imbalanced conditions.

0 citationsRead paper

Explainable post-training bias mitigation with distribution-based fairness metrics

Apr 01, 2025

This paper addresses the challenge of post-training fairness adjustment without model retraining. Methodologically, it introduces a novel distribution-driven differentiable fairness constraint optimization framework, designs an interpretable global bias metric family compatible with diverse models (e.g., gradient-boosted trees), and establishes a continuous, controllable post-processing paradigm for fairness calibration. Technically, it integrates distribution alignment constraints, interpretability regularization, and bias calibration. Evaluated on multiple benchmark datasets, the approach maintains high predictive accuracy while improving demographic parity (DP) and equalized odds (EO) by 30–50% over state-of-the-art post-processing methods. Moreover, it enables global bias attribution with human-interpretable explanations, offering both quantitative fairness enhancement and qualitative insight into bias sources.

0 citationsRead paper

Agentic AI Systems Applied to tasks in Financial Services: Modeling and model risk management crews

Feb 08, 2025

To address the challenge of poor integration between model development and model risk management (MRM) in financial services, this paper proposes the first multi-agent collaborative framework jointly optimized for financial modeling and MRM. Methodologically, it orchestrates two specialized agent teams—leveraging large language models (LLMs) for task decomposition, domain-knowledge-enhanced prompting, structured documentation generation, and automated model validation—to enable end-to-end autonomous collaboration across data exploration, feature engineering, model training, regulatory compliance review, reproducibility verification, and conceptual soundness assessment. The key contribution lies in the first deep coupling of modeling and MRM within a unified multi-agent architecture, ensuring both model performance and stringent regulatory adherence. Experiments on credit card fraud detection, credit approval, and portfolio credit risk modeling demonstrate significant improvements in modeling efficiency, interpretability, and regulatory compliance.

0 citationsRead paper