Abstract:

It is well known that classifiers learnt with a limited number of labeled samples are generally not robust for video concept detection. On the other hand, it is expensive and time-consuming to collect labeled training samples. Recently, many transfer learning (or domain adaptation) methods have been proposed to learn robust classifiers (referred to as target classifiers) with only a limited number of labeled patterns from the target domain by leveraging a large amount of labeled training data from other domains (referred to as auxiliary/source domains).

In the first part of the talk, I will introduce our recent multiple source domain adaptation method Domain Adaptation Machine (DAM), which leverages a set of pre-computed classifiers (referred to as auxiliary/source classifiers) independently learned with the labeled patterns from multiple source domains. Specifically, we introduce a new data dependent regularizer based on a smoothness assumption into Least-Squares SVM (LS-SVM), which enforces that the target classifier shares similar decision values with the auxiliary classifiers from relevant source domains on the unlabeled patterns of the target domain. Comprehensive experiments on the challenging TRECVID 2005 corpus demonstrate that DAM outperforms existing multiple source domain adaptation methods for video concept detection in terms of effectiveness and efficiency.

In the second part of the talk, I will briefly introduce our ongoing research projects which are related to internet vision, event recognition, and biometrics. 

Biography：

Dong Xu is currently an assistant professor at Nanyang Technological University in Singapore. He received the B.Eng. and PhD degrees from the Electronic Engineering and Information Science Department, University of Science and Technology of China, in 2001 and 2005, respectively. During his PhD study, he worked at Microsoft Research Asia and The Chinese University of Hong Kong for more than two years. He also worked at Columbia University for one year as a postdoctoral research scientist. His research focuses on new theories, algorithms and systems for intelligent processing and understanding of visual data such as images and videos.

He has published more than 35 papers in top venues including T-PAMI, T-IP, T-CSVT, CVPR, ACM MM, ICML, and IJCAI. His publications have been cited in Google Scholar more than 700 times, and his H-Index in Google Scholar is 15. Since 2008, he has received over S$2.8M in research grant funding, from Singapore National Research Foundation (NRF), A*STAR, Ministry of Education (MOE), and Microsoft Research Asia (MSRA). Ph.D. students in his research group were awarded the prestigious MSRA Fellowship Awards in 2008 and 2009.

He is an associate editor of Neurocomputing (Elsevier) and he is an Editorial Board Member of Journal of Multimedia (Academy). He has served as the guest editor of three special issues on video and event analysis in IEEE Transactions on T-CSVT, CVIU and PRL. He has also served as the Workshop Co-Chair of The ACM SIGMM Workshop on Social Media and the IEEE ICME Workshop on Visual Content Identification and Search, a Track Chair of IEEE International Conference on Multimedia & Expo (ICME) 2009, and a Theme Chair of Pacific-Rim Symposium on Image and Video Technology (PSIVT) 2009.