MSc Thesis Proposal Announcement by Mehrdad Sheikhjaberi:"Enhanced Defending Techniques Against Membership Inference Attacks"

 

Date: Tuesday October 18, 2022 

Time:  11:00am – 12:00 pm 

Location: Essex Hall, Room 122

Reminder: Two-part attendance required: Part I (Scan the QR code, fill in online form) and Part II (sign in sheet). *Make sure you are in the room at least 5-10 minutes BEFORE the presentation starts – once the presentation has begun, latecomers will not be admitted. 

Nowadays, because of the wide usage of machine learning models, privacy concerns about information leakage have been arising. There are different information leakage attacks, like membership inference attacks. In the membership inference attack, attackers try to detect whether someone’s data has been used during the model training. This will reveal sensitive information about people.  

 

Subsequently, we need to have mitigation techniques in order to use machine learning models safely. Moreover, the main challenge of the mitigation process is the tradeoff between model privacy and utility. We are planning to propose enhanced defending methods against membership inference attacks via Jacobian matrix and Entropy metric. 

 

Keywords: Machine Learning, Membership Inference Attack, Knowledge Transfer, Jacobian Matrix, Entropy 

Internal Reader: Dr. Mahdi Firoozjaei      

External Reader: Dr. Mohamed Belalia    

Advisor: Dr. Dima Alhadidi