Task 007: Distributed Learning and Inference/Exploiting Structure in Machine Learning to Create End-To-End Statistical-computing Platforms
|Event Date:||March 12, 2020|
|School or Program:||Electrical and Computer Engineering
Murat Ozatay, Princeton University
Co-design of machine-learning algorithms for control and perception with emerging technologies for structure-preservation in sensor data
It is now widely recognized that structure in data can be substantially exploited to enhance the performance of machine learning. Our recent research has noted that the real-world presents rich structure, which can be exploited in perception tasks through sensors that preserve such structure. We have proposed Physically-Integrated (PI) sensing as a technology that structures data around the (inter)actions and states of physical objects, through embedded, distributed, and form-fitting sensing directly coupled with objects. Large-Area Electronics (LAE), which can achieve large and conformal form factors, has been explored as a technology for achieving this. This talk will describe recent work pushing LAE to the giga-Hertz regime, opening up its use for wireless sensing. This leads to immersive wireless systems that can be integrated in walls and surfaces, and having extremely large apertures. Such apertures enable synthesis of rich radiation patterns for spatial addressing of density distributed PI sensors. However, they also required high-dimensional control over the many radiating elements, yet with limited analytical models for control. This talk presents that devices-to-algorithmic co-design of such systems, including: (1) the design and demonstration of multi-giga-Hertz wireless systems based on LAE; (2) use of deep-learning models for antenna control; (3) the integration of antenna controllers in complete perception pipelines, leveraging vision sensing for both selective sensor acquisition and sensor fusion with the acquired data.
Can Wu Bio:
Can Wu received the B.S. degree in microelectronic engineering from Tsinghua University, Beijing, China, in 2013, and the M.A. degree in electrical engineering from Princeton University, Princeton, NJ, USA, in 2016, where he is currently pursuing the Ph.D. degree. His research interests include thin-film circuit and CMOS IC hybrid system design for sensing applications.
Murat Ozatay Bio:
Murat Ozatay received the B.Sc. degree in electrical and electronics engineering from Middle East Technical University, Ankara, Turkey in 2015, and the M.A. degree in electrical engineering from Princeton University, Princeton, NJ in 2017, where he is currently pursuing his Ph.D. degree. He is a member of Verma Lab. His research focuses on bringing together algorithms and insights for learning with technologies and systems for advanced sensing. His primary research interests include machine learning, artificial intelligence, Internet-of-Things, and the design of VLSI systems.