Anonymisation for Time-Series Human Activity Data

Wearable motion data identifies the person wearing the sensor almost as reliably as it describes what they are doing. This work uses generative adversarial networks to strip identifying signal from IMU time series while preserving the information that downstream human activity recognition depends on — the trade-off between privacy and utility, measured on both axes.

Based on my M.Sc. thesis at TU Dortmund University.

Read the paper