MSc Thesis Proposal Announcement of Saroj Dayal:"Comparison of Membership Inference Attacks in Federated Learning "

 

Date: Wednesday November 9, 2022 

Time: 3:00pm – 4:00pm 

Location: Essex Hall, Room 122 
Reminder: Two-part attendance mandatory, arrive 5-10 minutes prior to event starting - LATECOMERS WILL NOT BE ADMITTED once the presentation has begun. 

 

Federated Learning received a lot of interest in its privacy protection feature. Federated Learning models are vulnerable to several inference attacks, like membership inference attack. In a membership inference attack, an attacker attacks the federated learning model to identify whether specific data samples have been used during the model training. Federated Learning models must be secured, especially during training, to preserve the privacy of the training datasets and to reduce information leakage. 

We compared two membership inference attacks in a federated learning environment and checked the effectiveness of the countermeasures on them. Additionally, we show through experiments which attack is more efficient with countermeasures while maintaining a comparable level of model accuracy. 

 

Keywords: Federated Learning, Membership Inference Attack, Jacobian Matrix 

 

Internal Reader: Dr. Shafaq Khan              

External Reader: Dr. Jagdish Pathak         

Advisor: Dr. Dima Alhadidi