[Seminar] Towards Safe Utilization of Machine Learning in Computing
Friday, October 18, 2024
11:00 am - 12:00 pm
Speaker
Dr.
Jianyi
Yang
Assistant
Professor
of
Computer
Science
University
of
Houston
Location
PGH 232
Abstract
Abstract: Artificial Intelligence (AI) is increasingly powering critical applications across various domains. To meet the rapidly-growing computing demand due to data processing, training and inference of AI, Machine Learning (ML)-based methods are proposed for workload scheduling and resource allocation in computing infrastructures. However, the use of ML in these environments raises significant safety concerns. For instance, an unsafe deployment of an ML policy for data center workload scheduling could potentially lead to energy emergencies. To ensure the safe use of ML in computing, we propose robustified machine learning algorithms to tackle several key online computing resource management problems. We provide the worst-case performance for these problems, offering the insights that can guide the design of safe ML policies. In this talk, I will first address continuous decision processes and introduce an algorithm that ensures the safety of reinforcement learning in data center management. Next, we will explore an online budgeted matching problem which models the matching process between virtual machines (VMs) and physical servers. I will present safe design framework for utilizing ML in the matching scenario based on the proven results. Finally, I will outline the research directions in the development of safe ML algorithms for modern intelligent systems.
About the Speaker
Jianyi Yang is an Assistant Professor in the Department of Computer Science at the University of Houston. Prior to that, he was a visitor at Caltech and a visiting researcher at UC Riverside. He received his PhD degree in 2023 at UC Riverside. His research interests include trustworthy AI and AI for efficient systems. His research has been published in top venues such as NeurIPS, ICML, SIGMETRICS, INFOCOM, and AAAI.
