Finger Bending Biometrics (Smart Glove)
Problem being addressed - Most behavioral biometrics for typing authentication rely on observable outcomes of motion
- keystroke timings or coarse hand movement using IMU sensors
- rather than the underlying biomechanics of fingers that generate those actions.
Why simpler approaches weren’t enough - Prior work using keystroke dynamics captures when keys are pressed but ignores how fingers move between presses.
- Wrist-worn sensors improve coverage but still treat the hand as a rigid unit, missing fine-grained finger articulation.
- As a result, these approaches collapse rich motor behavior into low-dimensional signals, limiting discriminative power.
Why human behavior + sensing mattered - Human dexterity is expressed at the finger level: subtle differences in bend, flex, and coordination emerge even when users type the same content.
- These micro dynamics are involuntary and difficult to consciously mimic, making them attractive for behavioral authentication but only if they can be sensed directly.
Key research question - Can finger bending dynamics captured directly via flex sensors serve as a reliable and discriminative biometric signal during typing?
- Rather than optimizing classifiers on existing signals, this work identifies a missing sensing layer in typing biometrics and introduces finger-level biomechanics as a new behavioral modality.
Power Side-Channel Video Inference (Public Charging Hubs)
Problem being addressed - Public USB charging hubs are widely assumed to be passive infrastructure that pose little privacy risk beyond malware-based attacks.
Why simpler threat models weren't enough - Most charging-related security concerns focus on data-line compromise (e.g., juice jacking).
- These models overlook the fact that power delivery itself is an observable signal
- one that can be monitored without modifying the phone or user behavior.
Why human behavior + sensing mattered - Video consumption is tightly coupled to human preferences, beliefs, and emotional states.
- Meanwhile, modern displays dynamically modulate brightness and color content, creating time-varying power signatures.
- When these two facts intersect, a seemingly benign sensor
- a power meter can become a privacy-invasive inference tool.
Key research question - Can power measurements taken at a public charging hub be used to infer which video a user is watching on their phone?
- This work reframes charging infrastructure as a side-channel sensor, exposing a gap between perceived safety and actual information leakage driven by human media consumption behavior.
Gamification as a Data Collection and Attack Vector (Wearables)
Problem being addressed - High-quality behavioral data collection for wearable systems is expensive, time-consuming, and dependent on sustained participant engagement.
Why simpler approaches weren't enough - Traditional lab-based protocols are monotonous and fail to scale, especially for machine learning models that require large, diverse datasets.
- Simply asking users to repeat actions often produces fatigue-driven artifacts that distort natural behavior.
Why human behavior + sensing mattered - Games naturally exploit reward, curiosity, and flow core aspects of human psychology.
- When coupled with wearable sensors, they can elicit authentic, repeated motor behavior without explicit user awareness of data collection.
Key research questions - Does gamified interaction preserve the behavioral patterns needed for wearable authentication?
- If so, could the same mechanism be exploited by an adversary to extract sensitive behavioral secrets?
- This work treats gamification not just as a UX tool, but as a behavioral amplifier
- capable of both enabling scalable sensing and introducing new privacy risks.
