Abstracts Communications and the Arts

Communication-Efficient and Robust Federated Learning for Heterogeneous Data

by Md Palash Uddin

This thesis investigated different aspects of information theory and optimization of conventional machine learning to best fit into the federated learning paradigm. The aims were to mitigate the key challenges (communication overhead, data heterogeneity, and lack of robustness) of federated learning. The developed methods in the thesis were successful to alleviate the aforementioned challenges.