Title: Energy-efficient Decentralized Optimization and Learning
Speaker: Prof. Zhi Tian, George Mason University, USA
Date/Time: August 29, 2020 / 9:10AM - 10:00AM EST
Abstract: In decentralized learning, a number of distributed nodes collaboratively carry out a common learning task in an autonomous manner, without sharing their private local raw data and often in the absence of centralized task coordination. In such a big-data paradigm, communication has become a common bottleneck in implementing efficient parallel and distributed algorithms, due to high latency and limited bandwidth of distributed networks. An ideal decentralized algorithm is expected to reach the optimal solution with minimal communication and computation costs. Nevertheless, the communication-computation tradeoff is essential. This talk presents some recent results on the design and analysis of energy-efficient schemes for decentralized consensus optimization through event-triggered control and communication censoring. These communication-saving strategies are illustrated via several optimization and learning problems of broad applications.
Bio: Prof. Zhi Tian has been a Professor in the Electrical and Computer Engineering Department of George Mason University since 2015. Prior to that, she was on the faculty of Michigan Technological University from 2000 to 2014, and served as a Program Director at the National Science Foundation from 2012 to 2014. Her research interests lie in statistical signal processing, wireless communications, and decentralized network optimization and machine learning. She is an IEEE Fellow. She is Member-at-Large of the IEEE Signal Processing Society Board of Governors. She was General Co-Chair of the IEEE GlobalSIP Conference in 2016. She served as an IEEE Distinguished Lecturer, and Associate Editor for the IEEE Transactions on Wireless Communications and IEEE Transactions on Signal Processing. She received the IEEE Communications Society TCCN Publication Award in 2018.
Title: Meta Learning in Edge Networks
Speaker: Prof. Junshan Zhang, Arizona State University, USA
Date/Time: August 29, 2020 / 10:10AM - 11:00AM EST
Abstract: Many IoT applications demand intelligent decisions in a real-time manner. The necessity of real-time edge intelligence dictates that decision making takes place right here right now at the network edge. Since an edge node often has a limited amount of data and is constrained with computational resources, we advocate meta learning to achieve edge intelligence. We first propose a federated meta-learning framework where multiple edge nodes collaboratively train the meta-learning model such that the obtained model can be quickly adapted by the target edge node, using its local data, to achieve real-time edge intelligence. We analyze the convergence of the federated meta-learning algorithm and examine the learning performance of the meta-trained model at the target edge. Next, we explore meta-reinforcement learning (RL) based system identification in a linear time-varying (LTV) system, which finds applications in many IoT edge networks. To address this open problem, we propose an episodic block model for the LTV system where the model parameters remain constant within each block but change from block to block. Built on this model, meta-RL based system identification consists of two main steps, namely off-line meta-learning and online adaptation. We carry out a comprehensive non-asymptotic analysis of the performance of meta-RL based system identification with correlated samples. Specifically, to deal with the technical challenges induced by strong sample correlation and small sample sizes, we devise a new two-scale small-ball approach to analyze the performance of off-line meta-RL, and then quantify the finite time error bound for online adaptation.
Bio: Junshan Zhang received his Ph.D. degree from the School of ECE at Purdue University in 2000. He joined the School of ECEE at Arizona State University in August 2000, where he has been Fulton Chair Professor since 2015. His research interests fall in the general field of information networks and data science, including communication networks, machine learning for Internet of Things (IoT), Fog/edge Computing, optimization/control of cyber physical systems, smart grid. He is a Fellow of the IEEE, and a recipient of the ONR Young Investigator Award in 2005 and the NSF CAREER award in 2003. His papers have won a few awards, including the Best Student paper at WiOPT 2018, the Kenneth C. Sevcik Outstanding Student Paper Award of ACM SIGMETRICS 2016, the Best Paper Runner-up Award of IEEE INFOCOM 2009 and IEEE INFOCOM 2014, and the Best Paper Award at IEEE ICC 2008 and ICC 2017. Building on his research findings, he co-founded Smartiply Inc in 2015, a Fog Computing startup company delivering boosted network connectivity and embedded artificial intelligence. He was TPC co-chair for a few major conferences in computer networks, including IEEE INFOCOM 2012 and ACM MOBIHOC 2015. He was general chair for ACM/IEEE SEC 2017 and WiOPT 2016.