报告题目：Using Relationships between Data and within Data for Ordinal Regression and Text Spotting
In machine learning, the independent assumption is always employed for classification. Data from the same class is assumed under an unknown distribution. The relationships between data and within data are not always used explicitly. In this talk, the speaker will discuss how to use the relationships between data and within data for ordinal regression and text spotting. In the first part of this talk, a deep ordinal regression algorithm will be presented. It encodes the ordinal data relationship through triplets of instances from different categories to train a deep neural network. In addition to encoding the data relationship, the triplet formation also serves as data argumentation. Thus, the network can be trained on small dataset. In the second part, the relationships between ordinal data are encoded in the constraints of an optimization formulation. Then, an unconstrained formulation is derived from the constrained optimization formation such that the standard backpropagation can be used to train deep neural network. In the last part of this talk, the speaker will show how to encode object label information in a deep neural network for text spotting, which is also called text detection. These works have been published in IJCAI, CVPR and ECCV. They are partially supported by National Research Foundation, Rolls-Royce and BAE Systems.
Dr. Adams Wai Kin Kong (江伟健) received the Ph.D. degree from the University of Waterloo, Canada. Currently, he is an associate professor at the Nanyang Technological University, Singapore. His papers have been published in leading journals and conferences in his research areas, including TPAMI, TIP, TIFS, TCSVT, Pattern Recognition, CVPR, EECV, IJCAI, ICB and BTAS. One of his papers was selected as a spotlight paper by TPAMI and another was selected as Honorable Mention by Pattern Recognition. With his students, he received best student paper awards in The IEEE Fifth International Conference on Biometrics: Theory, Applications and Systems (BTAS), 2012 and IEEE International Conference on Bioinformatics and Bioengineering, 2013. Dr. Kong received a number of awards from Singapore government for his cooperation with Singapore Police Force and the best reviewer award in BTAS 2016. In the summer of 2008, he served as an expert witness to the U.S. Department of Justice for a case of child sexual abuse. Currently, he is serving as an associate editor for IEEE Transactions on Information Forensics and Security. He has developed ten patents. His research interests include pattern recognition, image processing, biometrics, and forensics.