This project focuses on developing a collaborative machine learning system using federated learning. The system allows multiple participants to train a shared model without sharing raw data, preserving the privacy of individual datasets.
Study federated learning concepts and collaborative ML frameworks.
Identify privacy risks in centralized model training.
Design a system that allows distributed participants to train local models.
Implement aggregation of model updates without accessing raw data.
Apply encryption or secure aggregation techniques for updates.
Test system performance with simulated multi-client scenarios.
Evaluate model accuracy versus privacy protection.
Monitor communication efficiency and latency during training.
Document system scalability and security advantages.
Analyze potential attacks and propose mitigation strategies.
Prepare a final report comparing federated learning to centralized approaches.
