Bachelor and Master Theses

Background

Federated learning (FL) is a decentralized machine learning paradigm that enables multiple clients to collaboratively train a global model without exchanging raw data. Sensitive information is therefore kept local, which enhances data privacy. However, the decentralized nature of FL systems introduces new security challenges that concern, among others, model aggregation, data transmission, or clients that participate to perform attacks such as model poisoning.

Objective

This thesis aims to investigate possible secure architectural patterns for FL systems that stand to known security attacks and threats and satisfy the corresponding security requirements by reflecting on the trade-off necessary to make a FL system secure.

Tasks

The student is therefore requested, but not limited to

• Identify some attacks to FL systems and define the security goals/requirements to counter these attacks/threats
• Identify and evaluate secure architectural patterns of FL systems in the literature that satisfy the identified security requirement
• Study and discuss the impact of secure architecture on fundamental properties of FL systems, such as performance.

Requirements engineering, software architecture, AI & Machine Learning, Federated learning

This thesis can be assigned to one or two students. This thesis can be adapted for a bachelor thesis.