This project develops methods for protecting sensitive health data while improving fairness and accountability in AI systems. The research spans ECG re-identification, adaptive anonymization, synthetic data generation, causal fairness, and memorization risks in large language models.

Selected Publications

• Linkage Attacks Expose Identity Risks in Public ECG Data Sharing
• REACT: Reinforcement Learning-Based Adaptive ECG Anonymization and Privacy Threat Mitigation
• TransECG: Leveraging Transformers for Explainable ECG Re-identification Risk Analysis
• FairTabGen: Unifying Counterfactual and Causal Fairness in Synthetic Tabular Data Generation
• FairCauseSyn: Towards Causally Fair LLM-Augmented Synthetic Data Generation
• Skewed Memorization in Large Language Models: Quantification and Decomposition