Title of the report：Toward Cooperative Federated Learning over Heterogeneous Wireless Networks
Presenter：Seyyedali Hosseinalipour (Ali Alipour), University at Buffalo-SUNY, Assistant Professor
Time of the report：2023-May-16 (Tuesday) 10:00 am-11:30 am (Beijing Time)
Place of the report： 腾讯会议ID：608-542-903
Abstract of the report：Federated learning (FL) has been promoted as a popular distributed machine learning technique over wireless devices. We aim to integrate cooperation among the devices via device-to-device (D2D) communications into the conventional learning architecture of FL. We show device cooperation can compensate for the heterogeneity of devices in terms of data distributions and communication/computation resources. We consider conducting cooperative FL in an environment with time-varying data and propose the notion of concept drift. We then introduce idle times in between local model training rounds of cooperative FL to achieve resource savings, and optimize them with respect to concept drift. We obtain a set of convergence bounds on non-convex loss functions to describe the performance of trained machine learning models in cooperative FL and propose the notions of cold vs. warmed up model and model inertia. We finally formulate a general network optimization problem to optimize a tradeoff between machine learning model performance vs. resource consumption and solve it through a non-convex optimization technique based on Geometric Programming. Through numerical results, we reveal a series of observations on idle times, concept drift, and the performance of the trained machine learning model.
Biography of the presenter：Seyyedali Hosseinalipour (Ali Alipour) received the B.S. degree in electrical engineering from Amirkabir University of Technology, Tehran, Iran, in 2015 with high honor and top-rank recognition. He then received the M.S. and Ph.D. degrees in electrical engineering from North Carolina State University, NC, USA, in 2017 and 2020, respectively. He was the recipient of the ECE Doctoral Scholar of the Year Award (2020) and ECE Distinguished Dissertation Award (2021) at North Carolina State University. He was a postdoctoral researcher at Purdue University, IN, USA from 2020 to 2022. He is currently an assistant professor at the Department of Electrical Engineering at the University at Buffalo (SUNY).
He has served as the TPC Co-Chair of workshops related to distributed machine learning and edge computing held in conjunction with IEEE INFOCOM 2021&2023, IEEE GLOBECOM 2021, IEEE ICC 2021, IEEE/CVF CVPR 2023, IEEE MSN 2021-2023, IEEE VTC 2023. His research interests include the analysis of modern wireless networks, synergies between machine learning methods and fog computing systems, distributed machine learning, and network optimization.
Photo of the presenter:
Inviter：Dr. Minghui Liwang, Assistant professor, Dept. of Info. Commun. Eng.