Toward Privacy Preservation in Federated Learning: A Framework Integrating Client-Side Shuffling and Model...

Area of Science:

• Artificial Intelligence
• Machine Learning
• Cybersecurity

Background:

• Federated learning (FL) trains models collaboratively without sharing raw data, but faces challenges from client heterogeneity and privacy risks.
• Client heterogeneity impacts training efficiency and convergence speed.
• Inference attacks pose a threat during model parameter transmission.

Purpose of the Study:

• To propose an efficient and privacy-preserving federated learning framework (CSCP-Fed).
• To address challenges of client heterogeneity and enhance secure communication in FL.
• To improve training efficiency, convergence speed, and model generalization.

Main Methods:

• Implemented client-side shuffling and compressed-model techniques.
• Integrated differential privacy (DP) with client-side shuffling for secure aggregation.
• Developed an asymmetric-encryption protocol for secure data transmission.
• Introduced hybrid-weighted attention aggregation and compressed-model-driven client selection.

Main Results:

• CSCP-Fed effectively protects privacy without central entities.
• Reduced global loss by 15%-38% and communication overhead by 19.3%-44.3%.
• Accelerated convergence by 10%-41% and improved learning accuracy by 4%-13%.

Conclusions:

• CSCP-Fed offers an efficient and privacy-preserving solution for federated learning.
• The framework successfully mitigates heterogeneity impacts and enhances security.
• CSCP-Fed demonstrates significant improvements over traditional federated learning methods.