Institute of Information Science, Academia Sinica

Events

Print

Press Ctrl+P to print from browser

TIGP (AIoT) -- In-Memory Deep Learning via Emerging Memory Technologies

:::

TIGP (AIoT) -- In-Memory Deep Learning via Emerging Memory Technologies

  • LecturerDr. Hsiang-Yun Cheng (Research Center for Information Technology Innovation, Academia Sinica)
    Host: TIGP (AIoT)
  • Time2022-09-30 (Fri.) 14:00 – 16:00
  • LocationAuditorium106 at IIS new Building
Abstract
Deep neural networks (DNNs) have grown in prominence in recent years. Their intensive computing and memory demands introduce performance and energy efficiency challenges to the underlying processing hardware. In particular, the intensive data movements between CPU and memory incur severe performance degradation and energy penalties. One promising solution is to exploit the computing-in-memory capability of emerging memory technologies to build a revolutionary memory-centric computing architecture for deep learning. Nevertheless, its development is still in the early stage, and several design challenges need to be solved.
In this talk, I will introduce how emerging memory technologies enable energy-efficient DNN inference beneficial for AIoT devices. I will also present our recent studies that aim to overcome the new design challenges for such in-memory computing systems. At the end of the talk, I will share my vision for several future research directions.
 
BIO
Hsiang-Yun Cheng is currently an Assistant Research Fellow of the Research Center for Information Technology Innovation (CITI) at Academia Sinica. She received her B.S. and M.S. degree in Computer Science and Information Engineering from National Taiwan University and her Ph.D. degree in Computer Science and Engineering from Pennsylvania State University. Her research falls primarily in the field of computer architecture, with an emphasis on memory system design and domain-specific acceleration. She is especially interested in exploiting emerging technologies and the characteristics of modern applications to design energy-efficient computing systems.