Machine learning for attacking gesture-based phone unlocking project

• Main Author: Ng, Jun Hao
• Other Authors: Thomas Peyrin
• Format: Final Year Project
• Language: English
• Published: Nanyang Technological University 2021
• Subjects: Science::Mathematics
• Online Access: https://hdl.handle.net/10356/148511
• Institution: Nanyang Technological University

Nowadays, a smartphone contains numerous instruments, such as the accelerometer, gyroscope and proximity sensors, or the classical microphone or camera. While these instruments are the basis of the smartphone functionalities, they represent a potential security vulnerability if an attacker can have access to the data produced by them when the user is entering his PIN code or some password. Indeed, much research have shown how one can recover such secret information with good accuracy using state-of-the-art machine learning and deep learning algorithms when having access to these side-channel data. For example, the orientation of the phone, the variation of light, the sound acquired when typing the password, are all information that slightly leak some information about your secret data. This represents a serious threat as some applications may have access to these sensors and users sometimes do not necessarily understand the consequences of a too-permissive restriction policy. There were also studies on new classes of side-channel information, such as pupil movements, to retrieve a user PIN or password. The goal of this project is to study and combine different types of side-channel information to come up with a better model that can recover secret information with a higher level of accuracy.