Prof.Xiaochun Cao

Speech Title:

"ILL-Posed" Computer Vision Tasks

Abstract:

Computer vision tasks range from the simple perspective projection matrix estimation in a traditional camera calibration application to the large-scale foundation model fitting in a contemporary object detection cloud service. One may solve most computer vision tasks through fitting functions mapping the dense, if not continuous due to quantization, visual input to a discrete and meaningful output space, including categories, bounding boxes, and depths. Due to the significant difference in cardinalities of the domain and codomain, these mapping functions fail to meet one of the three Hadamard criteria for being well-posed. In other words, the unstable computer vision solution does not depend continuously on the parameters or input data. Many researchers are trying their best to design or learn computer vision algorithms being sufficiently robust to complex perturbations such as occlusion, smoke, rain, and fog. There are also scholars looking for dedicated but powerful adversarial perturbations. Does there exist an invariant backdoor perturbation that is capable to push an arbitrary image across the decision boundary in a classification task? Are all perturbations adversarial? In this talk, I will introduce these questions our team is exploring and briefly outline some of the progress. However, much still remains unclear in spite of our efforts, and we reiterate that there might not have the answers we're looking for before AI undergoes a brand new paradigm shift.

Bio of lecturer:

Xiaochun Cao received the B.E. and M.E. degrees in computer science from Beihang University (BUAA), China, and the Ph.D. degree in computer science from the University of Central Florida, USA, with his dissertation nominated for the University Level Outstanding Dissertation Award. He is currently a Professor and the Dean of the School of Cyber Science and Technology, Shenzhen Campus, Sun Yat-sen University (SYSU). In 2004 and 2010, he was a recipient of the Piero Zamperoni Best Student Paper Award from the International Conference on Pattern Recognition. He was on the editorial boards of IEEE Transactions on Circuits and Systems for Video Technology and IEEE Transactions on Multimedia. He is on the editorial boards of IEEE Transactions on Pattern Analysis and Machine Intelligence and IEEE Transactions on Image Processing.

Back to Invited Speakers/Panelists home page ↵