EE Colloquium: Collaborative Machine Learning: A Pluralistic Approach with Provable Guarantees

Abstract: With the massive amount of data generated by the proliferation of mobile devices and internet of things (IoT), coupled with concerns over sharing private information, collaborative machine learning and the use of federated optimization is often crucial for deployment of large-scale machine learning.  Despite recent progress on federated optimization, our understanding of some fundamental aspects of these methods, required for characterizing their performance guarantees, is still in its infancy. In this talk we will introduce a pluralistic heterogeneous distributed optimization framework for collaborative machine learning where the main idea is to integrate the personalization into the training. We also discuss novel stochastic communication-efficient distributed algorithms with provable convergence rates for adaptively exploiting underlying computational resources and overcoming the curse of data heterogeneity. Finally, we establish generalization bounds for the proposed algorithm and elaborate on different implications such as intensive compatibility and game theoretic implications. 
 

Biography: Mehrdad Mahdavi has been an assistant professor of Computer Science and Engineering at EECS since 2018. His primary research lies at the interface of machine learning and optimization with a focus on developing theoretically principled and practically efficient algorithms for learning from massive datasets and complex domains.  He has won several awards including Top Cited Paper Award from the Journal of Applied Mathematics and Computation (Elsevier) in 2010 and the Mark Fulk Best Student Paper Award at the Conference on Learning Theory (COLT) in 2012.