报告题目：Superpixel Gridization for Fast Object Localization [Slides]
文章信息：Liang Li, Wei Feng*, Liang Wan, and Jiawan Zhang, Maximum Cohesive Grid of Superpixels for Fast Object Localization, IEEE Computer Society Conference on Computer Vision and Pattern Recognition (CVPR’2013).
报告简介：To pursue efficiency, accuracy and scalability in large-scale image analysis, which form the fundamental requirements of Big Data, many recent algorithms in computer vision and media computing are built upon superpixels. Compared to the regularly sampled image pixels, the most advantages of superpixels are their good accordance to object boundaries and the considerably reduced number of variables that need to be handled. Hence, superpixel-based approaches usually have much better efficiency and scalability than pixel-level methods, but with comparable or better accuracy. However, the significant irregularity in the topological connections of superpixels encumbers direct applicaton of some successful efficient techniques in pixel-level processing, such as integral image and efficient subwindow search (ESS) etc., to the superpixel-level. This indeed becomes one major bottleneck of superpixel-level image processing. In this work, we focus on optimally regularizing superpixels with arbitrary topological structures into the regular grid structures, based on the scheme of cascade dynamic programming (CDP). We will specifically discuss the generic model of superpixel gridization, feasible initialization and fast optimization methods for superpixel gridization. We will also explore the application of superpixel grid in fast object localization and segmentation. Super pixel gridization will help to promote the seamless transition from pixel-level image processing to superpixel-level processing, by unleashing the great potentials of superpixel-based approaches.
报告人简介：Wei Feng received the B.S. and M.Phil. degrees in Computer Science from Northwestern Polytechnical University, China, in 2000 and 2003 respectively, and the Ph.D. degree in Computer Science from City University of Hong Kong in 2008. From 2008 to 2010, he worked as research fellow at the Chinese University of Hong Kong and City University of Hong Kong, respectively. He is an associate professor in School of Computer Science and Technology and the director of Computer Vision and Media Computing Center of Tianjin University. His major research interest is media computing, specifically including general Markov Random Fields modeling, discrete/continuous energy minimization, image segmentation, semi-supervised clustering, structural authentication, and generic pattern recognition. He has published more than 50 academic papers, including TPAMI, IJCV, TIP, PR, CVPR and ICCV. He got the support of the Program for New Century Excellent Talents in University, China, in 2011.
报告题目：Extended Variational Inference for Non-Gaussian Statistical Models [Slides]
 Z. Ma, A.E. Teschendorff, A. Leijon, and J. Guo, “Variational Bayesian Matrix Factorization for Bounded Support Data”, IEEE Trans. on Pattern Analysis and Machine Intelligence (TPAMI), Volume 37, Issue 4, pp. 876 – 889, Apr. 2015.
 Z. Ma, A. Leijon, “Bayesian Estimation of Beta Mixture Models with Variational Inference”, IEEE Trans. on Pattern Analysis and Machine Intelligence (TPAMI), Vol. 33, pp. 2160 – 2173, Nov. 2011.
 Z. Ma, P. K. Rana, J. Taghia, M. Flierl, and A. Leijon, “Bayesian Estimation of Dirichlet Mixture Model with Variational Inference”, Pattern Recognition (PR), Volume 47, Issue 9, pp. 3143-3157, September 2014.
 J. Taghia, Z. Ma, A. Leijon, “Bayesian Estimation of the von-Mises Fisher Mixture Model with Variational Inference”, IEEE Trans. on Pattern Analysis and Machine Intelligence (TPAMI), Volume: 36, Issue9, pp. 1701-1715, September, 2014.
报告简介：Recent research demonstrate that the usage of non-Gaussian statistical models is advantageous in applications where the data are not Gaussian distributed. With conventionally applied model estimation methods, e.g., maximum likelihood estimation and Bayesian estimation, we cannot derive analytically tractable solution for non-Gaussian statistical models. In order to obtain closed-form solution, we extend the commonly used variational inference (VI) framework via lower-bound approximation, by utilizing convexity/relative convexity of the integrants in the non-Gaussian distributions. In this presentation, we introduce the principles of the extended variational inference (EVI) and demonstrate its advantages in non-Gaussian mixture models and bounded support matrix factorization. We also show the advantages of non-Gaussian statistical models in real life applications, such as speech coding, 3D depth map enhancement, and DNA methylation analysis. Here, we restrict our attention to the non-Gaussian distribution in the exponential family
报告人简介：Zhanyu Ma received the M. Eng degree in signal and information processing from Beijing University of Posts and Telecommunications (BUPT), China, and the Ph. D. degree in electrical engineering from Royal Institute of Technology (KTH), Sweden, in 2007 and 2011, respectively. He has been an associate professor at the Beijing University of Posts and Telecommunications, Beijing, China, since 2014. From 2012 to 2013, he has been a Postdoctoral research fellow in the School of Electrical Engineering, KTH, Sweden. His research interests include pattern recognition and machine learning fundamentals with a focus on applications in multimedia signal processing, data mining, biomedical signal processing, and bioinformatics.