PATN: Real-Time Privacy-Preserving Predictive Adversarial Transformation Network for Mobile Sensor Data
Privacy on the Fly: A Predictive Adversarial Transformation Network for Mobile Sensor Data
AI summary
PATN uses historical signals to pre-generate adversarial perturbations and applies them immediately when mobile sensor data is collected, achieving real-time privacy protection. On two datasets, the ASR of the privacy inference model is 40.11% and 44.65%, with accuracy close to random; EER rises from 8.30% and 7.56% to 41.65% and 46.22%, and ASR is better than the baseline by 16.16% and 31.96%, respectively.