Distinguished Lecture by Prof. Seokbum Ko, IEEE CAS, at ETH AI Center

Deep Learning Computation and Its Application

Seokbum Ko, Ph.D., P.Eng.

IEEE CASS Distinguished Lecturer (2024-2025)

Department Head and Professor, Dept. of Electrical and Computer Engineering

Professor, Div. of Biomedical Engineering

University of Saskatchewan, Canada

Abstract: Deep learning has the potential to deliver exceptional performance across various applications. However, implementing deep learning models in real-world scenarios comes at a high cost due to their intensive demands on both computation and memory. The speed at which results are obtained—known as latency—relies heavily on computational efficiency.

In this lecture, we will explore computer arithmetic tailored for deep learning. We will begin by examining the computational requirements of deep learning models and layers. Then, we will delve into notable computer arithmetic designs from the literature, showcasing innovative approaches. Finally, we will discuss emerging trends in deep learning computation, including the promising Posit number system. To bring these concepts to life, real-world deep learning applications will also be shared with the audience.