PROJECTS
REACT: Reinforcement Learning-Based Adaptive ECG Anonymization and Privacy Threat Mitigation
This project introduces REACT, a reinforcement learning framework that protects sensitive electrocardiogram (ECG) data against re-identification threats. By dynamically injecting noise through a multi-objective reward system, REACT minimizes adversarial attack success while preserving clinical diagnostic accuracy. Tested on the MIT-BIH dataset, the approach reduces attacker accuracy to near-random levels while maintaining strong utility for healthcare tasks. Analyses using Pareto fronts, entropy, and AUC confirm that REACT outperforms conventional anonymization methods, achieving a superior balance between privacy and usability. This work advances privacy-preserving data sharing for secure and trustworthy digital health.