Securing Federated Deep Learning
Atharva Haresh Saraf, Shaurya Talewar, Susanta Das, Khushbu Trivedi, Ahmed A. Elngar · 2024
Federated learning is a cutting-edge approach to machine learning, providing an efficient and better way to collaborate in learning while also protecting the privacy of individuals. It allows devices and servers to gain the ability to train models collaboratively without sharing their private information. This distributed approach gives various benefits, including improved privacy, scalability, security, and meeting legal requirements. Federated learning, while offering many benefits, has drawbacks including uneven data distribution across devices, which can skew model training. Communication costs rise when exchanging models, making large-scale deployment challenging. Privacy concerns and performance trade-offs can limit model accuracy. Finally, the distributed nature of federated learning requires complex infrastructure and resource management to ensure optimal operation. This chapter deeply examines the advantages of federated learning and also explores practical solutions to overcome its disadvantages.