Events

"Energy-Constrained Advanced Sensing for Vision, Audition and Bioengineering"

Minhao Yang, Postdoc Researcher, Columbia University

March 1st (Thursday), 10:00am
Harold Frank Hall (HFH), Rm. 4164 (ECE Conf. Rm.)


Sensing at the edge of cyber-physical systems is often constrained by the amount of energy available from battery and/or energy harvesting. The conventional sensing paradigm of signal acquisition and raw data transmission is challenged by the energy-consuming wireless connection. Signal-aware encoding and local signal processing at the sensor nodes can significantly reduce data rate, which not only lead to energy saving but also may benefit low-latency actuation.

In my presentation, I will speak about the silicon implementations of two advanced sensing paradigms. In computational sensing, I will first introduce the dynamic vision sensor, and then I will show how the encoding fidelity can be improved by the encoding mechanism of asynchronous delta modulation. Nonlinear event decoding that can help further event rate reduction will be discussed with the consideration of event-timing uncertainty. In inference sensing, I will first describe the silicon cochlea that performs power-efficient acoustic feature extraction with event output, and then a one microwatt deep-neural-network-based voice activity detector designed with the aid of a customized abstract silicon cochlea model will be elaborated. I will also briefly discuss the potential applications of these two sensing paradigms in large-scale neural recording and low-power visual recognition, and conclude the talk with my vision of the evolution of sensing in future.

About Minhao Yang:

Minhao Yang is a postdoc researcher in the VLSI lab at Columbia University. He obtained his Ph.D. from ETH Zurich, and M.Sc. and B.Sc. from Peking University. He received Early Postdoc Mobility Fellowship for his postdoc proposal from Swiss National Science Foundation. His research interests include integrated intelligent sensing, low-power event-driven vision and audition sensors, and spike coding.

Hosted by: Professor Jim Buckwalter