Title: Towards Computation- and Communication-Efficient Federated Learning
Speaker: Pranay Sharma, Research Scientist, Department of Electrical and Computer Engineering (ECE) at Carnegie Mellon University.
Day, Date, and Time: Monday, 20th November, 2023, 09:30 - 10:30 AM.(Click here to join )
Abstract: Modern machine learning (ML) systems rely on the data collected at the edge devices to power diverse applications like predictive typing, personalized recommendations, and real-time traffic updates. However, data privacy concerns and network bandwidth constraints preclude gathering the entire dataset at a central location for further processing. In the past few years, federated learning (FL) has emerged as a natural solution to this problem. The edge devices in FL maintain exclusive control of their data and in return, shoulder part of the computational load of the central server. The devices communicate with the server periodically to synchronize their models. Google and Apple have already deployed FL to improve GBoard and Siri.
In this talk, he will discuss his work addressing several challenges in FL. Despite extensive research over the past few years, the underlying optimization problems solved by most work are simple minimization. However, many ML applications, like GANs, robust learning, and reinforcement learning, can be modeled as min-max problems. He will first describe his work solving nonconvex min-max problems in a federated setting. In addition to achieving state-of-the-art theoretical computation-communication guarantees, this work (somewhat surprisingly) also improves the existing centralized methods. Next, he will also talk about his work on FL systems solving minimization problems, where he characterizes the fundamental limits of the computation and communication required by FL methods, and quantifies the impact of limited device participation, where only a small fraction of all the devices may be available at any time. Finally, he will conclude with some open problems he is excited about and his broader research vision.
Bio: Dr. Pranay Sharma is a research scientist in the Department of Electrical and Computer Engineering (ECE) at Carnegie Mellon University. Pranay received his Ph.D. in ECE from Syracuse University in 2021. He finished his B.Tech-M.Tech dual degree in Electrical Engineering from IIT Kanpur in 2013. His current research focuses broadly on distributed machine learning and optimization. Specifically, his research interests include federated learning, stochastic optimization, differential privacy, and reinforcement learning.