Purdue University Younghyun Kim 교수님의 세미나가 6월 15일에 진행될 예정입니다.

관심 있는 학생 여러분들의 많은 참석 바랍니다.

■일시: 2026. 6. 15 (월요일) 15:00 ~ 17:00

■장소: 반도체관 400102

■주제: “Time-Accuracy Scalable Compute for Energy-Efficient Machine Learning”

■연사: Purdue University 전자전기컴퓨터 공학과 교수

Younghyun Kim is an Associate Professor in the Elmore Family School of Electrical and Computer Engineering at Purdue University, leading the Networked and Embedded Intelligent Systems Lab, and a faculty member of the Institute for Chips and AI, Center for Education and Research in Information Assurance and Security (CERIAS), Applied AI Research Center (AARC) at Purdue. Before joining Purdue in 2024, he was with the University of Wisconsin-Madison from 2016 to 2023. He was a Postdoctoral Research Assistant at Purdue University from 2013 to 2016. He received his Ph.D. degree in Electrical Engineering and Computer Science in 2013 and B.S. degree (highest honor) in Computer Science and Engineering in 2007, both from Seoul National University. His Ph.D. dissertation won the EDAA Outstanding Dissertation Award (2013). He is a recipient of the NSF CAREER Award (2019), Meta Research Faculty Research Award (2021), and other awards for innovative publications, designs, and demonstrations. He is a Co-PI of the NSF National AI Institute for Edge AI supported by NSF and DOH, and his research is supported by various federal agencies and corporations including NSF, USDA, Intel, and Meta. He is also collaborating with universities and industry in Korea, especially on research and education in semiconductors. His research interests include energy-quality scalable computing, edge/applied/physical AI, cyber-physical systems, and security and privacy for embedded computing systems.

■ABSTRACT

This talk presents time-accuracy scalable computing as a path toward energy-efficient machine learning beyond traditional semiconductor scaling. It highlights how approximate computing expands the design space by trading accuracy for improved performance and energy efficiency, while addressing key embedded-system challenges such as input-dependent quality variation, non-compute energy overhead, and severe resource constraints. The talk introduces several techniques, including low-latency stochastic computing, quality-configurable approximate arithmetic, approximate sensing and data transfer, and application-limited neural network design. Together, these approaches demonstrate how dynamic accuracy control and full-system optimization can substantially reduce latency, power, and energy while maintaining acceptable machine-learning accuracy.

■HOST: 고종환교수 (전자전기컴퓨터공학과)