Invited Speaker---Dr. Sang C. Lee


Daegu Gyoengbuk Institute of Scienc & Technology, South Korea

Biography: Sang C. Lee received the B.S., M.S., and Ph.D. degrees in electrical and electronics engineering from Pohang University of Science and Technology, Pohang, Korea, in 1994, 1996, and 2002, respectively. He was a Postdoctoral Researcher (supervisor: Romeo Ortega) with the Centre National de la Recherche Scientifique (CNRS), Supelec, France. Also, he was a project leader at the Samsung SDI Ltd. and Samsung Techwin to make a portable fuel cell and wireless sensor network, respectively. He is currently a Senior Researcher with the convergence research institute, Daegu-Gyeongbuk Institute of Science and Technology (DGIST), Daegu. His main research interests are in hybrid control and machine deep learning.

Speech Title: Improving discriminative feature learning for face recognition
Abstract: The model combining the center expansion cost function can learn the feature more efficiently than the model with the softmax cost function only. The features derived from the initial parameters are distributed randomly at the beginning of learning, so that the center of each class is equal to the center of all features. This means that all the features are gathered at the beginning of learning because of the error that reduces the variance of each class. Even if it is forced to derive the later distinguishing feature, the feature should be distributed in the dense space. However, by using the center expansion proposed in this paper, it is forced to prevent the feature from being derived from the center of all features, and also the features are gradually moved away from the center of all features to alleviate the dense problem. In this talk, the author proposes a forced model to derive more discriminative features from the model, and experiment with its effects. Experimental results show that the error rate can be further reduced by 28.67% compared with the conventional model. Future research will examine the performance of this paper using the Youtube faces database, which includes more people and images.