Auto-Key: Using Autoencoder to Speed Up Gait-based Key Generation in Body Area Networks
summary
Auto-Key uses a denoising autoencoder to predict chest accelerometer signals from distant wearables, raising bit agreement by 16.5% and making symmetric key generation 1.9× faster. Evaluation on a public dataset of 15 subjects with seven body locations shows predicted signals align much closer to chest readings, reducing mismatched bits. A transfer-learning tier retrains a general model with 50% less data and 88% less time. Passive, mimic and video attacks reach only ~62% bit agreement and fail to pair.
AI-written summary · about
chapters (11)
- 0:00Introduction
- 0:21Gatebased Key Generation
- 1:17Motivation
- 2:01Autoencoder
- 2:50Chain Model
- 3:35Evaluation
- 3:52Results
- 4:14Key Generation
- 4:57Baseline
- 5:12Security Analysis
- 5:22Conclusion
related talks
15mMaking Acoustic Side-Channel Attacks on Noisy Keyboards Viable with LLM-Assisted...
06mAudio Keyword Reconstruction from On-Device Motion Sensor Signals via Neural Frequency Unfolding
06mVoice In Ear: Spoofing-Resistant and Passphrase-Independent Body Sound Authentication
33mFor Smarter Authentication, We Might Need to Use ...
07mNot Quite Yourself Today: Behaviour-Based Continuous Authentication in IoT Environments
06mBlinKey: A Two-Factor User Authentication Method for Virtual Reality Devices
27mFast, Lean, and Accurate: Modeling Password Guessability Using Neural Networks
24mSecurity at your fingertips: Theory → Practice!