Advanced Lecture Series in Pattern Recognition
题 目 (TITLE)：Dark, Beyond Deep: A Paradigm Shift for Computer Vision
讲 座 人 (SPEAKER)：Prof. Song-Chun Zhu (University of California Los Angeles)
主 持 人 (CHAIR)： Prof. Cheng-Lin Liu
时 间 (TIME)：September 18 (Monday), 2017, 10:00 AM
地 点 (VENUE)：Academic lecture hall (3rd floor), Intelligence Building
In recent years, various vision tasks have been posed as classification problems optimized by end-to-end training using large examples. This deep learning paradigm is what I call “big-data for small-task”. In this talk, I will advocate a different paradigm: “small data for big tasks” i.e. using small examples to generalize to a range of tasks. Our human visual system is driven by a wide range of tasks in daily life, and visual representations, their precision, and computational mechanisms are all task-oriented. The design of objects and scenes are decided by the underlying Functionality and Physics, and human activities in videos are driven by underlying social Intents, Causality and Values. I call these underlying invisible FPICV entities as the “Dark Matter” which controls the visible geometry and appearance. I will show a number of examples and demonstrations.
Song-Chun Zhu received a Ph.D. degree from Harvard University in 1996. He is Professor of Statistics and Computer Science at UCLA. His work in computer vision received a number of honors, including the Marr Prize in 2003, the Marr Prize honorary nominations in 1999 and 2007. He received the Aggarwal prize from the IAPR in 2008, and the Helmholtz Test-of-time prize at ICCV 2013. He received NSF Career Award, and ONR Young Investigator Award in 2001. He is a Fellow of the IEEE since 2011. He serves the IEEE community as a General Chair for CVPR 2012 and 2019.