Data-driven Online Algorithms for Carbon-Intelligent Optimization

Abstract: The internet is a 24/7 service, it could be 24/7 carbon-free too. This is an ambitious goal that has been recently advocated by internet pioneer industries as part of their response to climate change. Achieving this goal, however, is challenging since it requires tackling a great deal of uncertainty in the environment for designing carbon-intelligent workload scheduling, load balancing, and energy procurement strategies. In this talk, we particularly focus on designing data-driven online algorithms for a carbon-intelligent energy procurement problem. Our solution design is motivated by the fact that rather than using hand-crafted worst-case optimized algorithms, practitioners prefer to optimize over a class of algorithms and tune the parameters of these algorithms to find an algorithm with improved performance in practice. Using this solution approach, we then present several energy procurement strategies that are provably robust against uncertainty and provide a design space for data-driven adaptation for improved practical performance. We also show how the theoretical foundations could be applied to other application domains.   

 

Bio: Mohammad Hajiesmaili is a Research Assistant Professor with the College of Information and Computer Sciences at UMass Amherst. Previously, he was a postdoc at Johns Hopkins University and The Chinese University of Hong Kong. He completed his Ph.D. and B.Sc. degrees from the University of Tehran and Sharif University of Technology. His research is at the intersection of computer science and energy. He received five best paper runner-ups from ACM eEnergy. His research is supported by NSF, Google, VMWare, and Adobe.