A research collaboration has found an efficient way to expand the collective behavior of swarming microrobots: Mixing different sizes of the micron-scale ‘bots enables them to self-organize into... Read more about Swarming microrobots self-organize into diverse patterns
Christina Delimitrou receives Microsoft Research Faculty Fellowship
Christina Delimitrou has received a Microsoft Research Faculty Fellowship, an award recognizing innovative, promising early-career professors who are exploring breakthrough, high-impact research.
Delimitrou is an assistant professor and the John and Norma Balen Sesquicentennial Faculty Fellow in electrical and computer engineering. She leads the Systems, Architecture, and Infrastructure Lab (SAIL) and her research interests include cloud computing, computer architecture and applied machine learning.
The Microsoft Research Faculty Fellowship recognizes promising new faculty, whose exceptional talent for research and innovation identifies them as emerging leaders in their fields. This year, 185 individuals were nominated and five were selected.
Delimitrou’s recent work focuses on leveraging machine learning to improve the performance predictability, resource efficiency and security of large-scale data centers. It’s had a significant impact on industry with several systems she has built being deployed in production cloud providers.
“In my group we have recently started a large undertaking to leverage machine learning in the design of a cloud-edge synergistic system,” said Delimitrou. “The system consists of a cluster of cloud servers and a swarm of edge devices, drones in our case, with the machine learning controller determining how work should be partitioned across cloud and edge to guarantee fault tolerant, responsive, and power-efficient computation.”
Delimitrou will use the funds provided by the fellowship to evaluate these systems at a larger scale, while also exploring more ambitious applications, such as digital agriculture and disaster recovery.
“A lot of unpredictability and inefficiency in current systems comes from empirical, heuristic-based designs,” Delimitrou said. “My approach has instead focused on applying data-driven, ML-based or analytical approaches to systems problems whose scale make previous empirical techniques ineffective.” She explained that while cloud systems are ideal candidates for this approach due to their scale, other systems can also benefit from data-driven techniques as hardware and software complexity increases.
For a long time, computer architects could rely on device technology delivering ever smaller, cheaper and faster transistors. There was no need to fundamentally rethink how systems were designed. Today that is no longer the case, which is why efforts to present novel architecture designs are getting significant attention from both academia and industry.
“Fellowships like this are unique,” Delimitrou said, “as they allow us to explore risky, but potentially highly impactful approaches.”