Ran Xu is now Dr. Ran Xu. He completed his PhD and graduated in December 2021. His thesis titled “Approximation for Streaming Video Analytics on Mobile Devices” broke new ground in how one can do approximation while meeting probabilistic guarantees on accuracy and latency.
Ran is now headed to NVIDIA in Santa Clara for working on their Deep Learning cuDNN and other software. Ran came to us straight from graduating from Tsinghua University and started in DCSL in Fall 2016.
During his thesis, Ran worked closely with professors outside DCSL as well. First, Prof. Sasa Misailovic at UIUC, then with Prof. Yin Li at U of Wisconsin at Madison and Prof. Somali Chaterji at Purdue (ABE). He also collaborated with Profs. Tarek Abdelzaher (UIUC), Prashant Shenoy (U Mass at Amherst), and Ramesh Govindan (U of Southern California) for a vision paper.
The main thesis that Ran’s work proved can be summarized as follows. This led to influential papers in Usenix ATC (2018), Sensys (2020), Intl. Symposium on Multimedia (2020), ACM Transactions on Sensor Networks (TOSN) (2021), and under review papers at Eurosys and CVPR.
Approximate streaming video analytic systems require content and contention aware characterization to achieve satisfied accuracy and latency performance to probabilistically meet accuracy and latency guarantees on resource-constrained mobile devices.