1. COMPSCI 720 Security for Smart-devices Tracking Mobile Web Users Through Motion Sensors: Attacks and Defenses [1] Harry Jackson hjac660
[1] Das, Anupam, Nikita Borisov, and Matthew Caesar. "Tracking mobile web users through motion sensors: Attacks and defenses." In Proceedings of the Network and Distributed System Security Symposium (NDSS), 2016.

2. Motivation Smart phone user privacy
How easy / simple it is to access a device’s sensors
Data from these sensors can be used to uniquely identify a user (AKA Sensor Fingerprinting)
Possible countermeasures for this invasion of privacy

3. Background
Fingerprinting has been used for identifying people for many years
Researchers have been attempting to imitate fingerprinting for many years
Things like radio-frequency, computer clocks, MAC headers, TCP/IP responses and more have been used as fingerprinting techniques

4. Background 2 Browser fingerprinting Sensor fingerprinting
Traces users across multiple websites / visits
Become more difficult in previous years due to privacy concerns
Sensor fingerprinting
Unique to smartphones
Uses the array of sensors that browsers have access to

5. Problem Das et al. aimed to:
Determine how accurate this sensor fingerprinting was at identifying users
Discuss and evaluate possible countermeasures
Themes of the paper very closely related to our course

6. Sensor Fingerprinting
Accelerometer
Uses capacitance to determine g-force exerted on it
Gyroscope
Uses the Coriolis effect to measure angular rate
A slight imperfection in either will result in different measurements, this can be used to uniquely identify devices as each phones defect is different

7. Sensor Fingerprinting 2
4 Streams of sensor data:

8. Sensor Fingerprinting – Results
They determined that using 70/100 features allowed for the highest accuracy
21 from accelerometer, 49 from gyroscope
44 of them being spectral features

9. Sensor Fingerprinting – Results 2
Sensitivity analysis:
Varying Number of Devices: Scaling from 10 to 93 devices the accuracy only decreased by 4%,
Varying training set: A 2:8 ratio still allowed for an accuracy of 98%
Varying temperatures: cold/hot environments did affect the accuracy, but was still mostly accurate
Temporal stability

10. Countermeasures – Calibration
The calibration errors that most phones posses allow for easier sensor fingerprinting.
Results:
Accelerometer accuracy dropped by 16-25%
When combined with gyroscope a high accuracy was still achieved

11. Countermeasures – Obfuscation
Obfuscation of the data returned from the sensors:
Uniform noise
Laplace noise
White noise

12. Summary Sensor fingerprinting Possible countermeasures
From the accelerometer and gyroscope
Very easy to access said data
Possible countermeasures
Calibration
Obfuscation

13. Issues / Improvements Accuracy of the Sensor Fingerprinting
The sample size was considerably small
Very controlled environment
Data obfuscations effect on the utility of the sensors
Only tested how it effected the step counter application
Das is doing further work in this area : ml