Advances in Federated Learning: Addressing Emerging Challenges and Expanding Frontiers through Optimization Theory

Federated Learning (FL) is an emerging machine learning paradigm that harnesses the computational power of user devices and utilizes distributed data while prioritizing user data privacy.

Despite its potential to revolutionize how we train and deploy AI models, FL faces unique challenges, such as non-homogeneity in data, communication overheads, and the need for personalized models.

This project aims to develop a comprehensive and systematic approach to address these challenges through knowledge transfer from optimization theory, focusing on three research thrusts: (1) Personalization via low-rank matrix factorization to develop effective and accurate personalized models; (2) Gradient compression and quantization t non-linear transformations inherent in non-Euclidean gradient methods to tackle communication overhead and privacy concerns without compromising from theoretical guarantees; and (3) Enhancing the Frank-Wolfe algorithm to solve constrained machine learning problems efficiently in federated settings.

By addressing these challenges, this research aims to significantly contribute to the development of efficient, robust, and privacy-preserving FL systems, with strong application potential in industries such as healthcare, finance, commerce, and public safety, where privacy and data security are critical concerns.