(0)

2018年中国矿业大学海外青年学者“越崎论坛”———计算机学院分论坛

来源:计算机科学与技术学院作者: 发布时间:2018-06-06 浏览次数:10

中国矿业大学学术报告

第二届计算机学科青年学者论坛暨首届“能源.革命.挑战”—中国矿业大学海外青年学者“越崎论坛”计算机科学与技术学院分论坛

  

间:201866日下午14:30

点:计算机学院B518

主办单位:计算机科学与技术学院

欢迎全校师生踊跃参加!

  

报 告 一:Quickly And Accurately Obtaining Image-based Localization

报 告 人: 卢国玉教授/博士(Guoyu Lu Professor/ Ph.D

: Rochester Institute of Technology

报告摘要:

Image-based localization is an essential complement to GPS localization. Current image-based localization methods are based on either 2D-to-3D or 3D-to-2D to find the correspondences, which ignore the real scene geometric attributes. We use a 3D model reconstructed by a short video as the query to realize 3D-to-3D localization under a multi-task point retrieval framework. Firstly, the use of a 3D model as the query enables us to efficiently select location candidates. Furthermore, the reconstruction of 3D model exploits the correlations among different images, based on the fact that images captured from different views for SfM share information through matching features. By exploring shared information (matching features) across multiple related tasks (images of the same scene captured from different views), the visual feature's view-invariance property can be improved in order to get to a higher point retrieval accuracy. More specifically, we use multi-task point retrieval framework to explore the relationship between descriptors and the 3D points, which extracts the discriminant points for more accurate 3D-to-3D correspondences retrieval. We further apply multi-task learning (MTL) retrieval approach on thermal images to prove that our MTL retrieval framework also provides superior performance for the thermal domain. This application is exceptionally helpful to cope with the localization problem in an environment with limited light sources.

个人简介:

Guoyu Lu is an Assistant Professor and PhD supervisor at the Chester F. Carlson Center for Imaging Science of Rochester Institute of Technology (RIT). Prior to joining RIT, he was a research scientist on autonomous driving at Ford Research and computer vision engineer at ESPN Advanced Technology Group. He finished my PhD and MS in Computer Science at the University of Delaware. Before coming to UD, he was in European Master in Informatics (EuMI) Erasmus Mundus program. He obtained Master degree in Computer Science at University of Trento and Master degree in Media Informatics at RWTH Aachen University. He also finished an academic visiting in Auckland University of Technology KEDRI group in 2012. He was a research intern at Siemens Corporate Research in Princeton and Bosch Research in Palo Alto. He finished my Bachelor degree in Software Engineering at Nanjing University of Posts & Telecommunications, with a minor in Business Administration and Management. He was the recipient of the most prestigious Frank Person Graduate Student Achievement Award at University of Delaware Computer Science Department and the winner of European Alliance for Innovation (EAI) Students Innovation Competition. He has published over 20 papers on international journals and conferences as the first author. He was the managing guest editor of the Springer Journal of Multimedia Tools and Applications. He was the area chair of the 4th Ford Global Control Conference and the organizer and leading program chair of the CVPR International Workshop on Visual Odometry and Computer Vision Applications Based on Location Clues.

报告摘要:

基于图像定位是GPS的重要补充。当前的2D图像和3D点云之间匹配的定位方法忽略了空间的几何信息。我们使用从短视频来重建出来的3D点云来进行定位从而实现了3D3D的定位。这个3D3D的定位是依据多任务学习来完成的。首先3D点云的使用使我们能够有效地选择候选地点。然后,基于SfM重建过程中的图片匹配过程,3D重建的过程使我们能够利用不同图片的关系。通过开发不同任务间的相关关系和共享信息,feature的针对不同视角的相关属性能够得到提升从而提高点对点的提取准确率。我们也同样使用热度图来处理夜间定位问题。

个人简介:

卢国玉现为罗切斯特理工学院(RITChester F. Carlson影像科学中心助理教授及博士生导师。 在加入RIT之前,他曾作为研究员任职于福特研究院和ESPN。 他在美国特拉华大学(UD)完成了计算机科学博士和硕士学位。 在加入UD之前,他曾就读于欧洲信息学硕士(EuMIErasmus Mundus项目,并获得了意大利特伦托大学计算机科学硕士学位和德国亚琛工业大学媒体信息学硕士学位。 他还在2012年对新西兰奥克兰理工大学KEDRI组进行学术访问。 他曾在西门子美国研究院和博世美国研究中心实习。 他在南京邮电大学完成了软件工程学士学位,并辅修工商管理。他曾获得特拉华大学计算机系最高奖Frank Person研究生成就奖和欧洲创新联盟意大利赛区第一名。他曾为国际期刊Multimedia Tools and Applications客座编辑,第四届福特全球控制大会领域主席,CVPR International Workshop on Visual Odometry & Computer Vision Applications Based on Location Clues 20172018年组织人及会议主席。

  

报 告 二:Balancing Cloud, Edge and Terminals: Cloud Gaming and Beyond

报 告 人: 蔡玮 博士(Ph.D

: The Universityof British Columbia

报告摘要:

To fulfill the rapidly increasing demand for computational resources in modern video games, cloud gaming has been proposed to leverage the power of cloud computing for delivering high-quality gaming experience to gamers anywhere and anytime. In this talk, I will start with my previous studies in cloud gaming, including cooperative video encoding, dynamic software partitioning, and data caching. The audience will gain the overview of cloud gaming research and familiarize themselves with the recent developments in this area. Also, I will further envision the future of computing convergence for user experience oriented applications, which integrates heterogeneous resources (e.g. cloud, edge and device-to-device paradigms) to provide supporting infrastructure in an elastic and multi-objective manner.

个人简介:

Dr. Wei Cai is a postdoctoral research fellow in Wireless Networks and Mobile Systems Laboratory at The University of British Columbia (UBC), Canada. He received Ph.D., M.Sc. and B.Eng. from UBC, Seoul National University (SNU) and Xiamen University (XMU) in 2016, 2011 and 2008, respectively. He has completed research visits at Academia Sinica (Taiwan), The Hong Kong Polytechnic University and National Institute of Informatics, Japan. He has published more than 20 first-author refereed articles and conference papers in the areas of cloud computing, interactive multimedia, and software systems. He received awards of the 2015 Chinese Government Award for the Outstanding Self-Financed Students Abroad, UBC Doctoral Four-Year-Fellowship, Brain Korea 21 Scholarship, and Excellent Student Scholarship from the Bank of China. He is also a recipient of the best paper awards from the CloudCom2014, the SmartComp2014, and the CloudComp2013. He also holds a faculty position at Kwantlen Polytechnic University, BC, Canada.

报告摘要:

为了应对现代视频游戏对于计算资源需求的极速增长,云游戏系统利用云计算资源向玩家提供随时随地的高质量游戏体验。本次报告将从共享视频编码、动态软件分拆、数据缓存等过往的云游戏研究出发,向听众简要介绍云游戏研究前沿和最新进展。由此,我们将展望未来面向用户体验应用程序的计算资源融合技术,包括整合云、边缘以及点对点网络等异构资源来提供弹性、多优化目标的基础设施。

个人简介:

蔡玮博士2008年毕业于厦门大学软件工程系,2011年获韩国首尔大学电气工程与计算机科学硕士,2016年获加拿大不列颠哥伦比亚大学(UBC)电气计算机工程学博士,现为UBC博士后研究员。求学期间曾赴日本国立情报学研究所、香港理工大学、台湾“中央研究院”等地从事访问研究工作,主要领域包括云计算、交互式多媒体以及软件系统,目前已在国际学术期刊和会议发表20余篇第一作者论文,曾荣获2015年度国家优秀自费留学生奖学金,CloudCom2014SmartComp2014以及CloudComp2013最佳论文奖等。蔡玮博士同时在加拿大昆特伦理工大学计算机科学与信息技术系担任教职。

  

报 告 三:Vision Recognition using Representation-based Classification Methods使用基于表示的分类方法进行视觉识别

报 告 人: 冯庆祥博士(Qingxiang Feng Ph.D

: 杜兰大学Tulane University

报告摘要:

Generally, vision recognition systems include feature exaction and classification processes. My research focuses on the classification process and intends to propose better classification methods (classifiers) for vision recognition. Recently, representation based classifiers have attracted the increasing attention of researchers. To obtain better performance, I proposed several better representation methods for classification. They are summarized as follows:

To obtain better projection representation for image recognition, this thesis proposes projection representation based classification (PRC) method. PRC is based on a new mathematical model denoting that the ideal projection of a sample point x on the hyper-space H may be gained by iteratively computing the projection of x on a line of hyper-space H with the proper strategy. Therefore, PRC is able to iteratively approximate the ideal representation of each subject for classification. Moreover, the discriminant PRC (DPRC) is further proposed to obtain the discriminant information by maximizing the ratio of the between-class reconstruction error over the within-class reconstruction error. Experimental results show that the proposed PRC and DPRC are effective on several vision recognition tasks.

To obtain better sparse representation for disease recognition, this thesis proposes the combined sparse representation (CSR) classifier. CSR minimizes sparse and the correlation structure of training set together for classification. Including the kernel concept, we propose the kernel combined sparse representation (KCSR) utilizing the high-dimensional nonlinear information instead of the linear information in CSR. Furthermore, considering the information of the training samples and the class center, we then propose the center-based kernel combined sparse representation (CKCSR) classifier. CKCSR uses the center-based kernel matrix to increase the center-based information for classification. The proposed classifiers have been evaluated by extensive experiments on several medical image databases.

To obtain better uncertainty representation, we propose regularized data uncertainty (RDU) and RDU coefficient (RDUC) classifiers. Compared to the traditional data uncertainty (DU) classifier, RDU and RDUC consider the importance of each sample for solving the minimum problem. Moreover, we propose kernel regularized data uncertainty (KRDU) classifier and kernel regularized data uncertainty coefficient (KRDUC) classifier to obtain the nonlinear information for data uncertainty. Extensive experiments on four benchmark action databases demonstrate that the proposed four classifiers achieve better recognition rates.

个人简介:

Dr. Qingxiang Feng got Ph.D from the Department of Computer Science at University of Macau in 2018. Now, He is working as Postdoctoral Fellow at Department of Biomedical Engineering of Tulane University. He has more than ten journal papers (first author) that been published and indexed by SCI, and has two top conference papers (1 for CVPR and 1 for AAAI). His research interesting includes computer vision and machine learning for medical analysis.

报告摘要:

视觉识别系统一般包括特征描述和分类过程。我的研究集中在分类过程上,希望提出更好的视觉识别分类方法(分类器)。近年来,基于表示的分类器越来越受到研究者的关注。为了获得更好的性能,我提出了几种更好的分类表示方法,总结如下:

为了获得更好的图像识别投影表示,本文提出了基于投影表示的分类方法PRC。该方法建立了一个新的数学模型,该模型表明,在超空间H上迭代计算x在超空间H里面线上的投影可以得到一个样本点x的理想投影。因此,PRC能够迭代地估计每个主题的理想表示,以便进行分类。此外,我们还提出了通过最大限度地提高类间重构误差与类内重构误差之比来获取判别信息的方法。实验结果表明,该方法对几种视觉识别任务有较好的效果。

为了获得更好的疾病识别的稀疏表示,本文提出了联合稀疏表示(CSR)分类器。CSR最小化了稀疏性和训练集的相关结构。在引入核概念的基础上,提出了利用高维非线性信息代替线性信息的核组合稀疏表示(KCSR)。并结合训练样本和类中心的信息,提出了基于中心的核联合稀疏表示(CKCSR)分类器。CKCSR使用基于中心的内核矩阵来增加基于中心的分类信息。通过对多个医学图像数据库的大量实验,对提出的分类器进行了评价。

为了获得更好的不确定性表示,我们提出了正则化的数据不确定性(RDU)RDUC (RDUC)分类器。与传统的数据不确定性(DU)分类器相比,RDURDUC考虑每个样本的重要性来解决最小问题。提出了核正则化数据不确定性分类器和核正则化数据不确定性系数分类器,获得了数据不确定性的非线性信息。在4个基准动作数据库上的大量实验表明,所提出的4个分类器能够获得更好的识别率。

个人简介:

冯庆祥博士于2018年在澳门大学计算机科学专业获得他的博士学位。现在他正在美国杜兰大学生物医药工程专业做博士后研究人员。他曾以第一作者发表超过10SCI期刊论文,另外还在顶级会议CVPRAAAI上面以第一作者分别发表一篇论文。他的研究兴趣包括计算机视觉和基于机器学习的医学分析。

  

报 告 四:Energy efficient wireless sensor networks

报 告 人: 黄艳秋博士(Dr.-Ing Yanqiu Huang

: University of Bremen

报告摘要:

Energy efficiency is the primary concern for almost any application of wireless sensor networks. We aim to firstly increase the intelligence of sensor nodes to enable them to understand data and then reduce the energy consumption for unnecessary activities. To further extend the network lifetime, we focus on developing new methods to distribute the tasks and balance nodes energy cost. The developed methods can not only reduce the energy cost, but also save bandwidth and mitigate big data problem, which is consistent with the goal of IoT, 5G and Cyber-physical systems.

个人简介:

Dr.-Ing Yanqiu Huang obtained her doctors degree at University of Bremen, Germany, where she is now working as a postdoctoral researcher. She obtained a young talent award in Germany in 2017 and funded by the central research development fund, Germany as a PI with around 1.23 million Yuan. She has published dozens of international journals and conference papers, and served as reviewers for several journals including Computer Networks,Computer Standards and Interfaces, ect.

报告摘要:

能源有效性对于无线传感器而言至关重要,很多科研工作都是围绕着这一主旨展开的。我们的目标是提高节点的智能性,让它们能够理解自己采集的数据,然后减少一些不必要的活动。从网络层面而言,考虑到节点的处理能力以及剩余能量的不同,我们需要对任务进行合理的分配以延长网络的生存时间。提出的算法在减少能耗的同时,也能够提高网络的吞吐量并减少大数据的问题。这和IoT5G以及Cyber-physical system的概念相吻合,使得我们的成果可以在这些大背景下进行应用。

个人简介:

黄艳秋博士在不莱梅大学取得博士学位,目前在该校从事博士后研究。 她在2017年获得杰出青年称号,并以PI的身份主持多个中央研究发展基金项目,合计约123万元人民币。在这期间,她发表十几篇国际会议及期刊论文,并担任相关期刊的审稿人,比如Computer Networks,Computer Standards and Interfaces等。

  

报 告 五:Trust Computing in Online Social Network

报 告 人: 刘冠峰博士(Ph.D

: The University of Queensland

报告摘要:

Online social networks have become the platforms for a variety of rich activities which include the selection of trustworthy targets via the recommendation of other participants. This requires mechanisms for the trust inference between non-adjacent participants, which are also expected to deliver trustworthy computation results. This talk will introduce the complex social network structures that take more information with influence on trust evaluation into account, and the methodologies and key problems included in trust computing in online social networks.

报告摘要:

在线社交网络已经成为信息发布共享,工作求职,产品推荐等活动的重要网络平台。由于参与者之间的信任影响着服务和产品的选择,信息的传播等。因此信任成为了支持人们网络社交的重要参考指标,具有重要的研究价值。本学术报告介绍社交网络的复杂结构以及社交网络中可信计算的关键方法和问题。

Bio:Dr.Guanfeng Liu received his Ph.D. degree in Computer Science from Macquarie University, Australia in 2013, and he received the nomination of John Makepeace Bennett Award - Australasian Distinguished Doctoral Dissertation. He is working as a Senior Lecturer in The University of Queensland, His research interests include graph mining and social networks. He has published over 40 papers in the most prestigious journals and conferences, such as AAAI, IJCAIICDE, CIKM, TKDE, TSC and ICWS.

个人简介:

刘冠峰,2013年毕业于澳大利亚Macquarie大学,获得计算机博士学位。博士论文获得当年澳大利亚最佳博士论文提名。目前为昆士兰大学高级讲师,主要从事可信计算、社交网络和图数据挖掘等领域的研究工作。曾参与多项澳洲科研基金项目研究。在AAAIIJCAITKDEICDETSCFGCSWWWJICWS等重要国际期刊与知名国际会议上发表论文近40篇。