Colloquium/Seminar

YearMonth
2019 Jan   Feb   Mar   Apr  
2018 Jan   Feb   Mar   Apr   May   Jun   Jul   Aug   Oct   Nov   Dec  
2017 Jan   Feb   Mar   Apr   May   Jun   Jul   Aug   Oct   Nov   Dec  
2016 Jan   Feb   Mar   Apr   May   Jun   Jul   Aug   Oct   Nov   Dec  
2015 Jan   Feb   Mar   Apr   May   Jun   Aug   Sep   Oct   Nov   Dec  
2014 Jan   Feb   Mar   Apr   May   Jun   Jul   Aug   Sep   Oct   Nov   Dec  
2013 Jan   Feb   Mar   Apr   May   Jun   Aug   Sep   Nov   Dec  
2012 Jan   Feb   Apr   May   Jun   Jul   Aug   Sep   Nov   Dec  
2011 Jan   Feb   Mar   Apr   May   Jun   Jul   Aug   Sep   Oct   Nov   Dec  
2010 Jan   Feb   Mar   Apr   May   Jun   Sep   Oct   Nov   Dec  
2009 Jan   Feb   Mar   Apr   May   Jun   Jul   Aug   Sep   Oct   Nov   Dec  
2008 Jan   Feb   Mar   Apr   May   Jun   Jul   Aug   Sep   Oct   Nov   Dec  
2007 Jan   Feb   Mar   Apr   May   Jun   Jul   Aug   Sep   Oct   Nov   Dec  
2006 Jan   Feb   Mar   Apr   May   Jun   Jul   Sep   Oct   Nov   Dec  
2005 Jan   Feb   Mar   Apr   May   Jun   Jul   Aug   Sep   Oct   Nov   Dec  
2004 Jan   Feb   Mar   Apr   May   Aug   Sep   Oct   Nov   Dec  

Event(s) on February 2019


  • Monday, 18th February, 2019

    Title: PX-MM Algorithm for Complicated Variance Components Model Estimation
    Speaker: Mr Zijian HUANG, Department of Mathematics, Hong Kong Baptist University, HKSAR, China
    Time/Place: 10:00  -  11:30
    FSC1217, Fong Shu Chuen Library, HSH Campus, Hong Kong Baptist University
    Abstract: Variance components model (mixed model) is a popular statistical approach to model hierarchical data. It is widely used to estimate associations, infer causal relationships, and do prediction. However, previously, the validation of mixed model in causal inference and prediction was restricted by small sample size. In order to avoid overfitting under a small sample size, statistical model is required to be simple but able to describe the general features of a data. But as data collection becoming more effective and more organisations willing to do it, sample size is not a restriction. Model can be more complicated to describe not only the general features of a data but more details, which is especially helpful in prediction and inference. But when model is complicated, previous methods of model estimation and prediction becomes ineffective. In this research, the idea of parameter-expansion (PX) is brought to accelerate the parameter estimation based on majorization-maximization (MM) algorithm under the circumstance of multi-response variables and multi-variance components.


  • Monday, 25th February, 2019

    Title: Semi-parametric Density Models
    Speaker: Professor Yuedong Wang, Department of Statistics and Applied Probability, University of California at Santa Barbara , California
    Time/Place: 11:00  -  12:00
    FSC1217, Fong Shu Chuen Library, HSH Campus, Hong Kong Baptist University
    Abstract: Maximum likelihood estimation within a parametric family and nonparametric estimation are two traditional approaches for density estimation. Sometimes it is advantageous to model some components of the density function parametrically while leaving other components unspecified. We propose estimation methods for a general semiparametric density model and develop computational procedures under different situations. We also present simulation results and real data examples.


  • Tuesday, 26th February, 2019

    Title: New recommended designs for screening either qualitative or quantitative factors
    Speaker: Mr Xiao KE, Department of Mathematics, Hong Kong Baptist University, HKSAR, China
    Time/Place: 09:30  -  11:00
    FSC1217, Fong Shu Chuen Library, HSH Campus, Hong Kong Baptist University
    Abstract: By the affine resolvable design theory, there are 68 non-isomorphic classes of symmetric orthogonal designs involving 13 factors with 3 levels and 27 runs. This work gives a comprehensive study of all these 68 non-isomorphic classes from the viewpoint of the uniformity criteria, generalized word-length pattern and Hamming distance pattern, which provides some interesting projection and level permutation behaviors of these classes. Selecting best projected level permuted subdesigns with $3leq kleq 13$ factors from all these 68 non-isomorphic classes is discussed via these three criteria with catalogues of best values. New recommended uniform minimum aberration and minimum Hamming distance designs are given for investigating either qualitative or quantitative $4leq kleq 13$ factors, which perform better than the existing recommended designs in literature and the existing uniform designs. A new efficient technique for detecting non-isomorphic designs is given via these three criteria. By using this new approach, in all projections into $1leq kleq 13$ factors we classify each class from these 68 classes to non-isomorphic subclasses and give the number of isomorphic designs in each subclass. Close relationships among these three criteria and lower bounds of the average uniformity criteria are given as benchmarks for selecting best designs.


  • Tuesday, 26th February, 2019

    Title: Generalized Power Methods for Group Synchronization Problems
    Speaker: Dr Man-Chung Yue, Imperial College Business School, Imperial College London, United Kingdom
    Time/Place: 15:00  -  16:00
    FSC1217, Fong Shu Chuen Library, HSH Campus, Hong Kong Baptist University
    Abstract: Group synchronization problems (GSPs) aim at recovering a collection of group elements based on their noisy pairwise comparisons and find a wide range of applications in areas such as molecular imaging, robotics and computer vision. Existing approaches to GSPs do not scale well and/or lack theoretical guarantees. In this talk, we focus on GSPs associated with two specific groups: the two-dimensional rotations SO(2) and the group of permutations. For each of these two groups, we propose to solve the GSPs by the generalized power methods and develop theoretical guarantees for the new approaches. Numerical experiments show that generalized power methods outperform existing approaches in terms of scalability, computation speed and recovery performance.


  • Wednesday, 27th February, 2019

    Title: Regularized Softmax for Semantic Image Segmentation
    Speaker: Fan JIA, Department of Mathematics, Hong Kong Baptist University, HKSAR, China
    Time/Place: 10:00  -  11:30
    FSC1217, Fong Shu Chuen Library, HSH Campus, Hong Kong Baptist University
    Abstract: Convolutional neural networks (CNNs) have achieved prominent performance in a series of image processing problems. Leading other traditional methods by a large margin, CNNs become the first choice for dense classification problems such as semantic segmentation. Despite higher accuracy and mIoU as CNNs achieved, no one has proposed an effective method to regularize the segmentation result. Thus the edge of segmentation result is often coarse, sometimes even serrated. Isolated and scattered small regions often appear in all kinds of CNN segmentation results. In this paper, we give the softmax activation function a new statistical interpretation with linear mixture model. In our method, the spatial regularization such as total variation (TV) can be easily integrated into CNN network and it can make the segmentation results more robust to noise when comparing with softmax. Meanwhile, we also introduce a training scheme to choose the regularization parameter λ instead of manually setting. We apply our proposed method to Unet, Segnet two networks and test them on WBC, CamVid two datasets. The results show that the regularized networks could achieve prominent regularization effect and better segmentation result.