Enabling Perception-Driven Optimization for Internet Video (via Zoom)

Abstract: The landscape of online applications has changed significantly, with the rise of interactive videos in daily remote learning and meetings, and the proliferation of AI applications that stream video sensor data to machine-learning models rather than human viewers, to make our cities safer and businesses more intelligent. The result is more video data and higher expectations of video quality, dramatically increasing the resource demands in network and compute. We argue that a driving contributor is that today's video systems try to optimize performance metrics (latency, buffering ratio, etc) everywhere. Instead, video systems should be built to directly optimize user experience (in terms of engagement of human viewers and accuracy of ML models). Through large-scale measurements and user studies, we show that users' sensitivity to traditional system performance metrics varies spatially and temporally more than previously believed. These findings have inspired new network system designs that improve the scale and user experience for human-centric applications (e.g., Internet videos) and AI-centric applications (e.g., video analytics).

Bio: Junchen Jiang is an Assistant Professor of Computer Science at the University of Chicago. He received his PhD degree from CMU in 2017 and his bachelor's degree from Tsinghua in 2011. His research interests are networked systems, multimedia systems, and their intersections with machine learning. He is a recipient of a Google Faculty Research Award,  NSF CAREER Award, and CMU Computer Science Doctoral Dissertation Award.