Zhiyuan graduate Yurong You published 3 papers on ICLR and CVPR

Zhiyuan graduate Yurong You published 3 papers on ICLR and CVPR Wenfang Fan 2022-08-28 1540

Autonomous cars "see" the world through sensors, and then use artificial neural networks to process the data from those sensors. They are different from humans, who have memories and become familiar with a path after a few trips, but for self-driving cars using artificial neural networks, the path is new every day. This can be a problem in bad weather, where sensors tend to be less reliable. To alleviate this problem, researchers from Cornell's Ann S. Bowers School of Computer and Information Science and School of Engineering published two research papers in CVPR 2022 and one paper in ICLR 22 that centered on the idea of creating "memories" for self-driving cars, and use those memories on subsequent drives.

Thesis 1 HINDSIGHT Is 20/20: A thought of noting Past Traversals to Aid 3D Perception. The study found that unsupervised historical data on the same location can greatly improve 3D sensing performance. "This information can be added as a feature to any LiDAR-based 3D object detector," says You. "The detector and SQuaSH can be trained jointly without any additional supervision or human annotation, which takes a lot of time and effort."

Thesis 2 Ithaca365: Dataset and Driving Perception under Repeated and Challenging Weather Conditions. Over 18 months, the team drove a car equipped with liDAR sensors on a 15-kilometer loop in and around Ithaca 40 times, collecting information along the way about the environment (highway, city, campus), weather (sunny, rainy, snowy) and different times of day.

Paper 3 Learning to Detect Mobile Objects from LiDAR Scans Without Labels takes another step forward, a method named Mobile Object Detection with Ephemerality and self-training (MODEST) is proposed. The authors let the self-driving car learn the entire perceptual process from scratch, and slowly it taught itself what other traffic participants are and what safety factors can be ignored. The algorithm can then reliably detect these objects, even on roads not included in the initial repeated traversal.

The first author of both paper 1 and 3 is Cornell doctoral student Yurong You (who also contributed to Paper 2). Yurong is a graduate of the Computer Science Direction (ACM Class) of Zhiyuan College. When he was interviewed at Zhiyuan College in his freshman year, Professor Alei Liang asked a question: "What is artificial intelligence?" ", and he firmly set foot on the road of artificial intelligence research ever after, to answer the question.

For Yurong, four years’ study at Zhiyuan, include sleepless nights and unforgettable excitement. In those four years, he experienced the freshman programming course computer test zero, also experienced the joy of compiler clearance, and he has always been committed to the study of artificial intelligence. In the summer vacation of his sophomore year, he joined the laboratory of Professor Tsewu Lu and started the research on computer vision and augmented learning. In the summer vacation of his junior year, he went to the AI laboratory of Stanford University and in the summer vacation of his senior year, he studied at Cornell University for scientific research internship in related fields. The scientific research internships at top overseas universities had broadened his scientific research horizon and introduced him to many interesting souls.

"Zhiyuan gave me the opportunity, courage and strength to pursue my dream," You said. I am very grateful to Mr. Cewu Lu, who led me on the road of scientific research and gave me confidence and unparalleled opportunities. I also miss my friends in ACM class. It was a pleasure to spend four years with you at Zhiyuan. " After his undergraduate studies, Yurong received a full PhD offer in Computer Science at Cornell University and the California Institute of Technology. In the end, he chose Cornell University, where he studied computer science and machine learning under Kilian Q. Weinberger, Professor of Computer Science, and Bharath Hariharan, Assistant Professor of Computer Science.