TITLE: Across-stack AI Systems Research in an Era of Cloud and Edge Computing

ABSTRACT:

As Moore’s law runs out of steam with CMOS technology approaching fundamental limits, heterogeneous systems become increasingly popular as a promising solution to solve the challenges of insatiable demand for high capacity, low latency and high energy efficiency computing. The recent booming of artificial intelligence (AI), machine learning and deep learning in particular, has further boosted such a demand for computing systems, including cloud computing and edge computing. This talk will discuss some of the research challenges resulting from these recent trends, and why a holistic AI systems research across the stack of software, middleware and heterogeneous hardware systems is of particular importance. I will show some exemplar research topics conducted at the IBM-ILLINOIS Center for Cognitive Computing Systems Research (c3sr.com), and try to shed some lights on the future directions of across stack AI Systems research.

BIO:

Dr. Jinjun Xiong is currently the Program Director for Cognitive Computing Systems Research at the IBM Thomas J. Watson Research Center. He co-founded and co-directs the IBM-Illinois Center for Cognitive Computing Systems Research (C3SR.com) with Prof. Wen-mei Hwu at UIUC. He was also a founding PI for the IBM Smarter Energy Research Institute (SERI) with deep collaboration with a number of large electrical utility companies worldwide. Prior to that, his technologies have been implemented inside IBM’s flagship EinsTimer/EinsStat tools, design and test methodologies used for designing multi-generations of IBM’s high-performance ASICs and Processors. He has published more than 100s of peer-reviewed papers in top AI conferences and systems conferences. His publication won five Best Paper Awards and eight Nominations for Best Paper Awards. He also led teams to win top awards for various international research competitions, including the recent championships for the DAC'19 Systems Design Contest on designing object detection neural network on low-power FPGA and GPU devices.