Abstract

Energy harvesting technology and ultra-low-power microcontrollers has led to the advent of tiny, battery-less devices, capable of sustainable, maintenance free operation. These devices can now run deep neural networks (DNN) locally, shifting intelligent decision making from the cloud to beyond the edge. However, as harvestable ambient energy is weak and unstable, battery-less devices operate intermittently, experiencing frequent power failures. This talk will first introduce the fundamental challenges with enabling deep learning on battery-less devices. Next, some of the work carried out in CITI's EMCLab to address these challenges will be highlighted, specifically introducing runtime software concepts for efficient intermittent DNN inference, and novel design automation frameworks that are able to design and optimize DNNs for intermittent systems.

Bio.

Dr. Hashan Roshantha Mendis is a Postdoctoral Scholar at the Research Center for Information Technology Innovation (CITI), Academia Sinica, Taiwan. He earned his MSc. and Engineering Doctorate (EngD.) degrees from the Department of Electronics and Department of Computer Science at the University of York in the United Kingdom in 2011 and 2017, respectively. Since 2018, he has been working in CITI's Embedded and Mobile Computing Lab. He has several years industrial experience, working as a software engineer in the digital TV, web and digital supply chain industries in UK, Taiwan and Singapore. His research interests broadly include embedded systems, with a focus on realizing Artificial Intelligence of Things (AIoT).

All faculty and students are welcome to join.