Distributed Optimization for Billions of Edge Devices
The field of distributed optimization has significantly matured in the last couple of decades. The interest in such techniques has been driven by managing resource allocation problems in information networks, and by machine learning problems with large amount of data to process. In this talk, the key techniques that have been developed will be surveyed first, highlighting the differences between distributed network utility maximization, the distributed sub-gradient method and the alternating direction method of multipliers. We will argue that the network abstractions used in developing these methods need to be advanced further to account for the hierarchical nature of actual networks, which becomes inevitable when the number of edge devices grows significantly. As examples we will discuss the federated learning network model and real-time pricing of electric power.
Prof. Anna Scaglione
Professor, Arizona State University on March 6, 2020 at 11:45 AM in EB2 1230
Anna Scaglione (M.Sc.'95, Ph.D. '99) is currently a professor in electrical and computer engineering at Arizona State University. She has applied her core expertise is in the broad area of statistical signal processing to various disciplines (data and network science, communication, control and energy delivery systems). One of her research focus is studying and enabling learning algorithm that are resileent and scale well when the data are distributed.
Dr. Scaglione was elected an IEEE fellow in 2011. She has has served in many capacities the Signal Processing, Communication and Control society over the years, and currently she is Deputy EiC for the IEEE Transactions of Control over Networked Systems. She received some awards: the 2000 IEEE Signal Processing Transactions Best Paper Award, the 2013, IEEE Donald G. Fink Prize Paper Award for the best review paper and her work with her student earned 2013 IEEE Signal Processing Society Young Author Best Paper Award (Lin Li). She is one of the 2019-2020 Distiguished Lecturers for the IEEE Signal Processing Society.
The Department of Electrical and Computer Engineering hosts a regularly scheduled seminar series with preeminent and leading reseachers in the US and the world, to help promote North Carolina as a center of innovation and knowledge and to ensure safeguarding its place of leading research.