On the Communication-Key Rate Region of Hierarchical Vector Linear Secure Aggregation.

Area of Science:

• Information Theory
• Federated Learning
• Network Security

Background:

• Federated learning faces challenges with heterogeneous data and task-specific aggregation.
• Secure aggregation is crucial for protecting user data in distributed systems.
• Hierarchical network architectures offer potential for improved efficiency and scalability.

Purpose of the Study:

• To investigate information-theoretic secure aggregation of linear functions in a two-hop hierarchical network.
• To define and characterize optimal communication rates for secure aggregation under privacy constraints.
• To develop a coding scheme that achieves these optimal rates.

Main Methods:

• Information-theoretic analysis using converse and achievability bounds.
• Modeling a system with an aggregation server, U relays, and UV users.
• Defining secure aggregation constraints for the server and relays.

Main Results:

• The study fully characterizes the optimal communication rates for secure aggregation in the defined network.
• An explicit linear coding scheme is proposed and shown to achieve the optimal rates.
• Hierarchical architectures enable optimal rates and reduce the server's masking burden.

Conclusions:

• Hierarchical networks provide a scalable solution for secure aggregation of authorized linear functions in federated learning.
• The proposed scheme balances security requirements with communication efficiency.
• This work advances the understanding of secure computation in distributed learning environments.