您好,欢迎来到华中科技大学自动化学院官网!

联系我们| English

当前位置: 首页 > 科学研究 > 学术报告 > 正文

学术报告第201736期

发布时间:2017-06-22 编辑:admin 来源:华科自动化学院

报告题目Learning Adaptation from Failure

人:Ying Wu教授美国西北大学

报告时间2017年6月26日15:00

报告地点南一楼西203室

摘要:A typical paradigm of an intelligent system is “off-line learning for on-line testing”, where the off-line learned model is applied blindly to all testing data regardless of their possible differences from the training data. Although this is a fundamental and widely adopted paradigm, it is based on an ideal assumption that the testing and training data are from the same distribution. However, in practice, the testing data are likely to be deviated, if not significantly different, from the training data. Thus, the learned model needs to be adapted to the individual testing data. However, the adaptation is very difficult, because no labeled data are available to re-train the model.

One example of our case study is object identification, a fundamental task of object recognition. There is a set of object images with known identities, called the gallery set. A test image, called a probe, needs to be identified against this gallery, by ranking order the gallery images. Typical solutions are to learn a faithful global metric off-line to cover the possible enormous variations in visual appearances, so that it can be directly applied to various probes for identity matching. Extensive research has indicated that such methods demand tremendous off-line learning over large sets of identity-labeled data. In contrast, we advocate a different paradigm that shifts part of learning to on-line, or learning adaptation, but with nominal computational costs. Here we assume that a global metric has been obtained, and we aim to learn an adaptive local metric for each individual input probe.

A major challenge here is that no positive training data is available for the probe anymore, as its identity is unknown. The interesting innovation in our new approach is to learn the adaptation only from negative data, or called “learning adaptation from failure”. Such negative data are the failure cases of the global metric, which can be easily obtained by different means. Given a probe, we only use such negative data to learn a strictly positive semi-definite dedicated local metric. We give in-depth theoretical analysis and justification of this new approach. We prove that it guarantees the reduction of the classification error asymptotically, and prove that it actually learns the optimal local metric to best appropriate the asymptotic case by a finite number of training data. Comparing to global metric learning, its computational cost is negligible. In addition, it is generally applicable, as it can be easily integrated on top of any global metric only to enhance its performance.

 

报告人简介:Dr. Ying Wu is full professor of Electrical Engineering and Computer Science at Northwestern University, Evanston, Illinois. He received his B.S. from Huazhong University of Science and Technology, Wuhan, China, in 1994, the M.S. from Tsinghua University, Beijing, China, in 1997, and the Ph.D. in electrical and computer engineering from the University of Illinois at Urbana-Champaign (UIUC), Urbana, Illinois, in 2001. In 2001, he joined the Department of Electrical and Computer Engineering at Northwestern University as an assistant professor. He was promoted to associate professor in 2007 and full professor in 2012. His current research interests include computer vision, image and video analysis, pattern recognition, machine learning, multimedia data mining, and human-computer interaction. He serves as associate editors for IEEE Transactions on Pattern Analysis and Machine Intelligence (IEEE T-PAMI), IEEE Transactions on Image Processing (IEEE T-IP), IEEE Transactions on Circuits and Systems for Video Technology (IEEE-TCSVT), SPIE Journal of Electronic Imaging (JEI), and IAPR Journal of Machine Vision and Applications (MVA). He serves as Program Chair and Area Chairs for top conferences including CVPR, ICCV, and ECCV. He received the Robert T. Chien Award at UIUC in 2001, and the NSF CAREER award in 2003. He is a Fellow of the IEEE.