- This event has passed.
Ph.D. Defense: Sohaib Ahmad
August 18, 2022 @ 1:30 pm - 2:30 pm EDT
Title: Privacy Leakages in Biometric Systems
Ph.D Candidate: Sohaib Ahmad
Major Advisor: Benjamin Fuller
Associate Advisors: Alexander Russell ,Caiwen Ding
Data/Time: Thursday, August 18th, 2022, 1:30 PM
Location: ITE 336
Meeting link : https://uconn-cmr.webex.com/uconn-cmr/j.php?MTID=me4046d4d3edb05de688a753dd3a49396
Password: eD3p2mc3YSZ
Join by phone: +1-415-655-0002
Meeting number: 2622 321 6428
Access code: 262 232 16428
Join by video system
Dial 26223216428@uconn-cmr.webex.com
You can also dial 173.243.2.68 and enter your meeting number.
Abstract:
A biometric is a biological trait used to identify and recognize humans. Biometrics are commonly used to authenticate users for sensitive applications. These applications can range from a user gaining access to a mobile phone or an employee entering a security sensitive facility. The process of recognizing humans from their biometrics involves many steps.
Segmentation is step one where biometric data is separated from non-biometric data. Feature extraction is step two where information is extracted from the biometric data to generate a template. This entire process is noisy due to a noisy biometric capturing process. We first study the entire iris recognition process which involves iris segmentation from an image and then feature extraction performed on the segmented iris. Most of our studies are performed using deep learning networks where we identify key components in the iris recognition process and adapt to use deep learning effectively. We show competitive accuracy figures for both segmentation and recognition.
Secondly, privacy issues arise when using deep learning networks on sensitive data such as biometric images. Specifically, we study two attacks on deep neural networks: membership inference attacks and model inversion attacks. These attacks leak private information about individuals in certain settings and open a secure system to more powerful spoofing attacks where adversaries can gain unauthenticated access to a secure system. We discuss both attacks and defenses in multiple settings and the effect privacy preserving legislations (such as the GDPR) have on these attacks.
