XR privacy research
configure-bs
A notebook-first research pipeline for privacy-preserving Beat Saber telemetry.
Even anonymized XR motion telemetry can retain enough behavioral signal to re-identify the person who generated it.
A five-stage pipeline: Unity + SteamVR telemetry capture, baseline re-identification measurement on raw motion, WGAN-GP synthetic generation, DP-SGD-style differential privacy constraints, and CKKS homomorphic-encryption analytics kept as a separate evaluation track.
This is the code behind my IEEE VR 2025 paper (doi: 10.1109/VRW66409.2025.00281): generative models used defensively, cutting re-identification risk from 100% to 0.04% while preserving analytical utility.
Key results
- Beat Saber / SteamVR telemetry preprocessing
- Raw-data identification baseline
- Synthetic-data privacy and utility evaluation
- Differential privacy experiments
- Homomorphic-encryption notebook direction
Tech stack
PythonTensorFlowWGAN-GPDifferential privacyCKKSJupyter
Project media
Evidence from the build.
