Zoom: https://illinois.zoom.us/j/83089271099?pwd=ZlRwQkw3bTU4bU5tdWxPbXM0bkNiZz09
Abstract:
Today’s cloud datacenters are tremendously complex yet prevalently used as the backbone for continuously evolving applications, such as microservices, serverless, and machine learning (ML). Various ML techniques have shown their strength in systems and networking domains. However, a substantial gap exists in directly applying ML advances to real-world cloud systems due to violations of a series of assumptions in practice.
This talk demonstrates an ML-centric cloud systems management framework that fundamentally integrates learned abstractions at its core to bridge this gap. By co-designing systems and ML algorithms, we can (1) optimize service-level objective (SLO) management with learned resource abstractions (OSDI’20), (2) support multi-tenancy with learned virtual agent abstractions (SoCC’22, NeurIPS’22), and (3) provide fast model adaptation with learned meta-learner abstractions (ATC’23, NeurIPS’23). Ultimately, I will outline my vision for making ML-centric systems highly accessible, efficient, and sustainable for the upcoming decade.
Bio:
Haoran Qiu is a PhD candidate at the Computer Science Department of the University of Illinois Urbana-Champaign, advised by Ravishankar Iyer. His research lies at the intersection of distributed systems, cloud computing, and machine learning, with a focus on building practical data-driven cloud management systems to optimize performance, resource, and energy efficiency. His work has been adopted in the open-source community and is deployed at IBM and Alibaba. His research appears in venues like OSDI, ATC, SoCC, EuroSys, and NeurIPS. He was selected as an ML and Systems Rising Star in 2023. He was a recipient of the Mavis Future Faculty Fellowship in 2022 and the Yunni & Maxine Pao Memorial Fellowship in 2021. During his PhD, he was a visiting researcher at IBM Research. Before UIUC, he obtained his BEng in Computer Science from the University of Hong Kong.
https://wiki.illinois.edu/wiki/display/grad/PILOT+SEMINAR
