Home > Seminars > Yingyan Lin - Energy-Efficient Machine Learning Systems for Edge and Cloud Computing

Yingyan Lin - Energy-Efficient Machine Learning Systems for Edge and Cloud Computing

Start:

9/20/2018 at 3:30PM

End:

9/20/2018 at 4:45PM

Location:

138 DeBartolo

Host:

College of Engineering close button
headerbottom

Yiyu Shi

Yiyu Shi

VIEW FULL PROFILE Email: yshi4@nd.edu
Phone: 574-631-6520
Website: http://www.nd.edu/~scl/index.html
Office: 325D Cushing

Affiliations

Department of Electrical Engineering Concurrent Associate Professor
College of Engineering Associate Professor
low-power design, three-dimensional integration, hardware security, renewable energy applications
Click for more information about Yiyu
574-631-6520
Add to calendar:
iCal vCal

Machine learning (ML) systems have been dramatically transforming the way we live and work by enhancing our ability to recognize, analyze, and classify the world around us. Current ML systems adopt either a centralized cloud computing or a distributed edge computing paradigm. In both paradigms, the excellent performance of ML algorithms comes with high complexity, leading to a huge energy challenge. In edge computing, energy efficiency is the primary design challenge, as edge devices have limited energy, computation and storage resources. This challenge is being exacerbated by the tremendous need to embed ML algorithms, such as deep neural networks (DNNs), for enabling local on-device learning and inference capabilities. On the other hand, in cloud computing, data transfer due to inter-chip, inter-board, inter-shelf and inter-rack communications (I/O interface) within data centers is one of the dominant energy costs. This will only intensify with the enormous demand of increased I/O bandwidth for high-performance computing in data centers.

In this talk, I will introduce holistic system-to-circuit approaches, that jointly optimize all design levels from system and architecture down to circuit/device, to address the energy challenge in both edge and cloud computing. First, I will present two techniques that can potentially enable on-device deployment of DNNs in edge computing by significantly reducing the energy consumption. Second, I will present the world’s first bit-error-rate optimal analog-to-digital converter based serial link receiver for realizing energy-efficient data transfer in cloud computing, demonstrating this with the implementation of a 4 Gb/s link in 90nm CMOS. Measurement results have shown that this technique provides a promising solution to the well-known I/O interface power bottleneck problem in data centers. Finally, I will share some exciting applications that our current research in machine learning systems for resource-constrained platforms are being applied to.

Seminar Speaker:

Yingyan Lin

Rice University

Yingyan Lin is an Assistant Professor in the Department of Electrical and Computer Engineering at Rice University. Her research focuses on energy-efficient machine learning systems for edge and cloud computing. She is leading the Efficient and Intelligent Computing (EIC) Lab at Rice, which is exploring algorithm-, architecture-, and circuit-level techniques to develop energy-efficient machine learning systems.

Yingyan Lin received a Ph.D. degree in electrical and computer engineering from the University of Illinois at Urbana-Champaign in 2017. She was the recipient of the 2nd place Best Student Paper Award at the 2016 IEEE International Workshop on Signal Processing Systems (SiPS 2016) and the 2016 Robert T. Chien Memorial Award for Excellence in Research, and selected as a Rising Star in EECS by the 2017 academic career workshop for women at Stanford University.