Finger Bending Biometrics — Smart Glove
Participants (N) - 55 users total
- 23 users: numeric PIN entry
- 32 users: alphanumeric password entry
Sessions / repetitions - Multiple entries per user per condition
- Repeated trials per authentication task to capture intra-user variability
Task constraints - Users typed:
- A 9-digit PIN on a laptop number pad
- A 10-character password on a full-sized keyboard
- Gloves worn consistently during typing to maintain sensor alignment
Variability intentionally captured - Differences between:
- Numeric vs alphanumeric typing
- One-glove vs two-glove configurations
- Finger-level variability, including non-typing fingers
- Population-level spread via per-user EER distributions
Design intent - Model how typing behavior varies within the same user, not just between users
- Stress-test the modality under realistic typing conditions rather than fixed scripts
Gamified Wearable Data Collection — Smartwatch IMU
Participants (N) - Multiple users (exact N varied by experiment)
Sessions / repetitions - Repeated pattern executions across:
- Traditional (explicit authentication) sessions
- Gamified interaction sessions
- Multiple repetitions per pattern per user
Task constraints - Users executed smartphone pattern-lock gestures in two modes:
- Direct authentication-style input
- Game-based interactions designed to mimic pattern trajectories
Variability intentionally captured - Behavioral drift between:
- Serious vs playful interaction contexts
- Conscious vs incidental execution of patterns
- Natural variability in wrist motion over repeated sessions
Design intent - Test whether data collected "in the wild" (via games) preserves biometric fidelity
- Compare controlled vs ecologically valid behavior rather than assuming equivalence
Power Side-Channel Video Inference — Charging Hubs
Participants / devices (N) - 5 smartphone models
- 100 distinct YouTube music videos
Sessions / repetitions - Multiple playback sessions per video
- Repeated measurements across configuration settings
Task constraints - Phones charged via instrumented charging hubs
- Videos played under controlled environmental settings
Variability intentionally captured - 16 brightness levels
- 16 volume levels
- Cross-device generalization (train/test on same vs different phone models)
- Temporal variation in video content and color dynamics
Design intent - Probe robustness of inference under realistic user-controlled settings
- Identify which variables materially affect attack success vs which do not
