孙剑 微软亚洲研究院

报告题目：Deep Image Understanding: From Generic Image to Face, Object to Pixel

报告内容：In this talk, I will briefly introduce our recent researches of generic image understanding at three levels: image classification, object detection, and semantic pixel labeling. The talk covers a number of topics: how to achieve the state of the art accuracy on image classification (e.g., surpassing human performance on ImageNet 1K challenge)? How to design a highly efficient object detection pipeline? How to exploit large scale but weakly-label data for semantic pixel labeling? At the end of the talk, I will also introduce a few recent, exciting compute vision projects in Microsoft Research.

讲者信息：Jian was born in Xian, China, which is home of The Terracotta Army. He received a B.S., M.S., and a Ph.D. degree from Xian Jiaotong University in 1997, 2000 and 2003, respectively. Immediately following, he joined Microsoft Research Asia, and has been working in the fields of computer vision and computer graphics, with particular interests in building real-world working systems. His current primary research interests are computational photography, face recognition, and deep-learning based image understanding.

Jian has published 70+ papers in the most prestigious conferences or journals (Google Scholar citation >10,000 times and H-index-42). He won CVPR (Computer Vision and Pattern Recognition) 2009 Best Paper Award, and was named one of the world's top 35 (TR35) young innovators by MIT Technology Review in 2010. He was an area chair of ICCV 2011/CVPR 2012/CVPR 2015 and paper committee member of SIGGRAPH 2011. Jian has filed 30+ US or International patents and many of his research results have been incorporated within Microsoft products, including Windows, Office, Bing, and Xbox. He is also an adjunct professor of Xian Jiaotong University and University of Science and Technology of China.

   　　田奇 美国德克萨斯大学圣安东尼奥分校

报告题目：Discriminative Visual Representations for Image Search and Classification

报告内容： The emergence of massive image data has urged effective and efficient techniques towards large-scale image search and classification. Generally, the introduction of local invariant features and the Bag-of-Words model has witnessed the development of both research fields. However, this model suffers from both the information loss during vector quantization and the deficiency in the descriptive power of various aspects of images. In the light of this problem, this talk will present several recent works in our group that focus on the discriminative representations in image search and classification. To be specific, I will first describe feature representations such as binary descriptors, region/part descriptors including CNN, as well as informative quantization. Then, in accordance to the descriptors, I will introduce a series of recently proposed indexing schemes aiming at reducing both memory and time cost, such as cross-index, cascade category-aware index, tensor-index, super-image index, etc. Third, I will present some exciting advances in discriminative feature fusion strategies, such as coupled multi-index, bayes merging, co-index, etc. Moreover, our recent works on post-processing such as ImageWeb, query-adaptive fusion will also be discussed. Last but not least, I will summarize our achievements in advancing the state-of-the-arts on benchmark datasets in image search and classification.

讲者简介： Qi Tian is currently a Full Professor in the Department of Computer Science, the University of Texas at San Antonio (UTSA). During 2008-2009, he took one-year Faculty Leave at Microsoft Research Asia (MSRA) in the Media Computing Group. He received his Ph.D. in ECE from University of Illinois at Urbana-Champaign (UIUC) in 2002 and his B.E and M.S degrees from Tsinghua University and Drexel University in 1992 and 1996, respectively, all from electronic engineering. Dr. Tian’s research interests focus on multimedia information retrieval and computer vision and published over 280 refereed journal and conference papers. He received the Best Paper Award in PCM 2013, ACM ICIMCS 2012 and MMM 2013, a Top 10% Paper Award in MMSP 2011, the Student Contest Paper Award in ICASSP 2006. His research projects are funded by NSF, ARO, DHS, Google, FXPAL, NEC, SALSI, CIAS, Akiira Media Systems, HP and UTSA. He received 2010 ACM Service Award. He is the Associate Editor of IEEE Transactions on Multimedia, IEEE Transactions on Circuits and Systems for Video Technology (TCSVT) and in the Editorial Board of Journal of Multimedia (JMM) and Journal of Machine Vision and Applications (MVA), he is also the Guest Editors of IEEE Transactions on Multimedia, Journal of Computer Vision and Image Understanding, etc. He is the Guest or Adjunct Professor in Institute of Computing Technology, Chinese Academy of Science, Xi’an Jiaotong University, University of Science and Technology of China (USTC), Zhejing University, Xidian University and a Chaired Professor in Tsinghua University.

   　　周志华 南京大学

报告题目：从AdaBoost到LDM

报告内容：本报告先介绍关于AdaBoost的一些理论研究结果，然后介绍在该结果启示下产生的机器学习算法设计新思想，以及随之而来的有效算法。

讲者简介：周志华，南京大学教授，ACM Distinguished Scientist，IEEE Fellow，IAPR Fellow，中国计算机学会会士。国家杰出青年基金获得者，长江学者特聘教授。主要从事机器学习、数据挖掘、模式识别等领域的研究工作，出版英文著作一部，获发明专利十余项，在一流国际期刊和顶级国际会议发表论文逾百篇，被引用万余次。先后担任十六种SCI(E)期刊编委，现任《Frontiers of Computer Science》执行主编，《中国科学：信息科学》等刊副主编，ACM TIST、IEEE TNNLS等刊副编辑。四十余次国际会议主席或领域主席，系列国际会议ACML发起人，现任ACML指导委员会主席，PAKDD、PRICAI指导委员会委员，IJCAI 2015顾问委员会委员及机器学习总主席，IEEE ICDM 2015程序委员会主席等。曾获国家自然科学二等奖、两次教育部自然科学一等奖、IEEE计算智能杰出青年成就奖、中国青年科技奖、微软教授奖、霍英东青教一等奖、12次国际期刊/会议论文、报告或竞赛奖等。已培养毕业博士生9名，硕士生20余名，学生获全国优博1次，CCF优博7次、江苏省优博/硕7次、微软学者7次等。

   　　王晓刚 香港中文大学

Tutorial: Deep Learning for Computer Vision

Part I: Introduction to Deep Learning

内容：Deep learning has become a major breakthrough in artificial intelligence and achieved amazing success on solving grand challenges in many fields including computer vision. Its success benefits from big training data and super parallel computational power emerging in recent years, as well as advanced model design and training strategies. In this part of tutorial, we will try to introduce deep learning and explain the magic behind it with layman terms. Through concrete examples of computer vision applications, we will focus on four key points about deep learning. (1) Different than traditional pattern recognition systems, which heavily rely on manually designed features, deep learning automatically learns hierarchical feature representations from data and disentangles hidden factors of input data through multi-level nonlinear mappings. (2) Different than existing pattern recognition systems which sequentially design or training their key components, deep learning is able to jointly optimize all the components and crate synergy through close interactions among them. (3) While most machine learning tools can be approximated with neural networks with shallow structures, for some tasks, the expressive power of deep models increases exponentially as their architectures go deep. (4) Benefitting the large learning capacity of deep models, we also recast some classical computer vision challenges as high-dimensional data transform problems and solve them from new perspectives.

讲者信息：王晓刚2011年于中国科技大学少年班获得学士学位,2004年于香港中文大学信息工程系获得硕士学位,2009年于美国麻省理工学院获计算机博士学位。自2009年至今任香港中文大学电子工程系助理教授。他于2012年获得香港杰出青年学者奖,香港中文大学青年学者奖。他是国际期刊Image and Visual Computing Journal的副主编,IEEE Transactions on Circuits and Systems for Video Technology的副主编, 2011年和2015年IEEE国际计算机视觉大会(ICCV) 的领域主席, 2014年欧洲计算机视觉大会(ECCV)的领域主席,2014年亚洲计算机视觉大会(ACCV)的领域主席。他的研究兴趣包括计算机视觉、深度学习、群体视频监控、物体检测、和人脸识别等等。

   　王乃岩 香港科技大学

Tutorial: Deep Learning for Computer Vision

Part II: Computer Vision Applications

内容：Despite its success in traditional tasks such as image classification and object detection, the power of deep learning has been fully released in the computer vision community. In the second part of the tutorial, we will introduce some emerging applications in video analysis, image caption generation, 3D estimation and reinforcement learning. At last, we will summarize some key factors that contribute to a successful application of deep learning, and provide a recipe on how to customize it.

讲者信息：Naiyan Wang is currently the final year PhD candidate in CSE department, HongKong University of Science and Technology. His supervisor is Prof. Dit-Yan Yeung. Before that, he got my BS degree from Zhejiang University, 2011 under the supervision of Prof. Zhihua Zhang. His research interest focuses on applying statistical computational model to real problems in computer vision and data mining. Currently, he mainly work on sparse representation, matrix factorization, deep learning, especially interested in the area of visual tracking, object detection, image classification and recommender system.

  　方璐 中国科学技术大学

报告题目：Deblurring Night Images: Utilizing Separable Kernel and Saturated Regions

报告内容：The blur effects caused by camera shake and object motion in scenes occur frequently in photography, producing disappointing blurry images with inevitable information loss, which becomes one of the most common reasons for discarding photographs. In this talk, I would like to present our recent work in deblurring night images, which is challenging due to the low contrast combined with heavy noise and saturated regions in night images. Unlike existing methods that discard saturated regions, we propose a novel scheme to deduce blur kernels from saturated regions via novel separable kernel representation and advanced algorithms. The beauty of our work is that it does not rely on whether the deblurring scheme can estimate a right kernel at the initial stage. While under the iterative initialization and blind deconvolution, our work provides the deblurring scheme a chance to break away from local minimum with the help of new initial blur kernel, even without changing the deblurring scheme.

讲者简介： Dr. Lu Fang is currently an Associate Professor in the Department of Electronic Engineering and Information Science, the University of Science and Technology of China (USTC). She was a Visiting Scholar at Northwestern University in 2010, a Visiting Professor in Microsoft Research Asia (MSRA) in 2013, and a Visiting Professor in Technical University of Munich (TUM) in 2015.

Dr. Fang received her Ph.D. in Electronic and Computer Engineering from HKUST in 2011, and B.E. in Electronic Engineering and Information Science from the USTC in 2007, respectively. Dr. Fang’s research interests include Subpixel Rendering, Computational Photography and Computer Vision. Dr. Fang has published over 50 IEEE international conference and IEEE journal publications, including IEEE SPM, IEEE TIP, IEEE TCSVT, IEEE TMM, IEEE CVPR etc. She used to be awarded for Humboldt Research Fellowship for Experienced Researchers, 10% paper award in ICIP 2014, Best Paper Candidate in ICME 2011. Dr. Fang has served as TC member in Multimedia Signal Processing Technical Committee (MMSP-TC) in IEEE Signal Processing Society.

　黄畅 百度（美国）

报告题目：Computer Vision at Baidu IDL

报告内容： In the past two years, Baidu IDL has invested a lot of talents and resources in developing fundamental cutting-edge computer vision technologies, in particular within the framework of deep learning. In this talk, I would like to briefly introduce our latest computer vision related research and products, demonstrating our achievements in object detection with fully convolutional network, OCR by sequence learning, and face recognition by metric learning. Future research directions will be highlighted and discussed in the end of the talk.

讲者简介：百度深度学习研究院（IDL）主任研发架构师/高级科学家，视觉图像技术负责人。本科、硕士以及博士毕业于清华大学计算机科学与技术系，先后就职于美国南加州大学和NEC美国研究院，于2012年加入百度美国研究院。长期从事计算机视觉、机器学习、模式识别和信息检索方面的研究，是相关学术界和工业界的知名专家，发表的论文被引用两千余次，拥有多项国际专利，开发的人脸检测识别技术被广泛用于电子数码产品和图像管理软件之中。自2012年夏天加入百度至今，他带领整个图像技术团队负责公司内各种图像核心技术的研发，产出了全网人脸图像搜索、PK大咖、全网相似图像搜索、自然场景文字识别、百度移动图像搜索等重要产品。

   　　姜育刚 复旦大学

报告题目：海量视频的内容分析与识别

报告内容： 在该报告中，我将分享复旦大学计算机学院视频分析团队近年来的部分研发成果。团队在视频特征表示、面向视频数据的深度学习方法、大规模视频分析基准数据集构建等方面取得了一些进展，发表了多篇顶级国际期刊、会议论文，产生了一定国际影响。在应用方面，团队开发的系统在企业和国家重大应用中做出了贡献，受到了上级单位表扬。

讲者简介： 姜育刚，香港城市大学博士、美国哥伦比亚大学博士后，2011年通过复旦大学人才引进回国工作，现为计算机科学技术学院副教授、博士生导师，视频大数据分析实验室负责人。研究领域为多媒体信息检索与计算机视觉，主要关注海量视频、图像内容分析与检索技术。至今在国际知名期刊和会议上发表论文70余篇。据Google Scholar，总被引用2700余次，单篇最高被引用近500次，H因子为25。目前主持国家自然科学基金、科技部863计划、上海市科委等资助的多个研究课题。

  　李武军 南京大学

报告题目：大数据哈希学习

报告内容： 海量性是大数据最重要的特点之一。数据的海量性将造成存储开销大、检索速度慢等问题。哈希学习（learning to hash）通过机器学习将数据表示成二进制哈希码的形式，能显著减少数据的存储开销。同时，基于哈希学习得到的哈希码可以构建索引机制，从而实现常数级别或者次线性级别的快速检索。因此，哈希学习于近年来发展成为大数据机器学习（简称大数据学习）中的一个研究热点。本报告将首先介绍哈希学习的应用背景与研究进展，然后详细讲解报告人自己最近的研究成果，最后对值得进一步探索的几个问题进行展望。

讲者简介： 李武军，博士，副教授，博士生导师。主要研究领域为人工智能、机器学习与数据挖掘。2010年9月至2013年12月于上海交通大学计算机科学与工程系从事教学与科研工作。2014年1月加入南京大学计算机科学与技术系。曾获Google奖教金、入选南京大学登峰人才支持计划（B层次）。发表论文30余篇，其中大部分发表在IEEE Transactions TKDE、AAAI、CVPR、ICML、IJCAI、NIPS、SIGIR等国际知名期刊和会议上。现任《Frontiers of Computer Science》青年副编辑，担任TPAMI、TNNLS、TKDE、TPDS、TCSVT、中国科学、科学通报、软件学报等多个国际和国内知名期刊的特邀评审人，并担任AAAI、ICML、IJCAI、NIPS、SIGKDD等多个国际知名会议的程序委员或者评审人。主持和参加了多项国家级课题研究，包括国家自然科学基金项目、863重大项目等。更多信息请参见个人主页

   　　李永杰 电子科技大学

报告题目：基于生物视觉机理的智能图像处理技术

报告内容： 人脑是已知最有效的生物智能系统。视觉是人类获取信息的最直接与最主要的手段。图像/视频作为视觉信息的重要载体，其内容分析与理解是开发新一代智能机器人、自动目标跟踪系统和实时人机交互界面等研究领域的核心技术之一。鉴于环境的复杂多变性和工程应用的苛刻性，传统的图像处理技术面临着巨大挑战。探索生物视觉的感知机理，建立以自适应机理为核心的视觉计算模型框架与分析技术，有望为复杂图像和视频的内容提取、精炼与解析，以及计算机视觉高层应用（如基于视觉传感的环境感知、理解与决策）等问题的解决提供一系列共性关键技术支撑。本报告将在概要介绍大脑视觉系统基本知识及国际前沿的基础上，重点介绍我们在近几年的相关研究成果，包括视网膜、初级视皮层等的计算模型及其在图像去噪、去雾、增强、亮度恢复（明/暗适应）、颜色恢复（颜色恒常）、轮廓检测等方面的应用。

讲者简介： 电子科技大学生命科学与技术学院（神经信息教育部重点实验室）教授、博士生导师、生物医学工程系主任。2004年在电子科技大学生命学院获生物医学工程专业博士学位。2007年入选教育部新世纪优秀人才。2009年9月至2010年9月在美国哥伦比亚大学神经科学系做访问学者。目前担任中国电子学会生物医学电子学分会副秘书长、中国计算机学会计算机视觉专业组委员、中国自动化学会生物控制论与生物医学工程专业委员会委员。迄今已在国际/国内期刊、国际会议上发表/录用学术论文60余篇。2008年以前主要从事生物智能计算及其在医学图像处理中的应用。近几年来主要从事生物视觉信息加工机理、计算模型及其在智能图像处理中的应用研究（如图像增强、亮度/颜色恒常、轮廓提取、显著区域检测等）。相关论文以第一作者或通讯作者发表在NeuroImage (IF=6.132), IEEE Trans. on PAMI, IEEE Trans on Image Processing, IEEE Trans on Biomedical Engineering等国际期刊，以及计算机视觉及模式识别领域的一流国际会议（CVPR, ICCV, ECCV等上面（其中含1篇ICCV’2013 oral paper）；申请中国发明专利17项（已授权10项）。更多信息参见课题组网站

   　　刘偲 中科院信工所

报告题目：Fashion Analysis: Parsing, Attribute Prediction and Retrieval

报告内容： In this talk, I will first briefly present an overview on the recent development in fashion analysis. Then I shall introduce in detail our several recent works in the fashion parsing, attribute prediction and retrieval. Firstly for fashion parsing, three works will be introduced. The first work utilizes the video context to solve the problem of insufficient training samples, and assist the fashion parsing. The other two works are deep learning based methods, including a parametric framework namely active template regression and a non-parametric framework called Matching-CNN. About fashion attribute prediction, I shall introduce a new framework to jointly estimate the fashion attribute and infer their corresponding semantic regions. Great performances have been achieved by the proposed methods in different aspects of fashion analysis.

讲者简介： Dr.SiLiu is now an Associate Professor in Institute of Information Engineering, Chinese Academy of Sciences. She used to be a Research Fellow at the Department of Electrical and Computer Engineering, National University of Sin-ga-pore (NUS).She obtained PhD degree from Institute of Automation, Chinese Academy of Sciences (CASIA) in 2012. She obtained Bachelor degree from Experimental Class of Beijing Institute of Technology (BIT).Her current research interests include attribute prediction, object detection andimage parsing. She is also interested in the applications, such as makeup and clothes recommendation, online product retrieval. She received the Best Paper Awards from ACM MM'13, Best Demo Awards from ACM MM'12.

   　　聂飞平 西北工业大学

报告题目：A Simple and General Algorithm for Various Norms Minimization or Maximization

报告内容： In many machine learning methods, we need to minimize or maximize certain norm functions, and how to efficiently optimize these functions is a fundamental and important issue. In this talk, I will first introduce a very simple algorithm to minimize a general function, and then demonstrate how it can be applied to minimize or to maximize various norm functions. The algorithm will find a global optimal solution if the problem is convex, and a local optimal solution if the problem is non-convex. The algorithm is very simple, general, and useful. For example, it can be used to minimize most robust loss functions for robust learning, and can also be used to maximize several norms for robust PCA/LDA, sparse PCA, etc.

讲者简介： 聂飞平，西北工业大学“光学影像分析与学习中心”教授、博士生导师。2009年于清华大学自动化系获博士学位，之后在美国德州大学阿灵顿分校担任研究助理教授，2015年入选中组部第六批青年千人计划并回国任教。主要研究方向为模式识别与机器学习算法及应用，已在国际顶级期刊和会议（PAMI、IJCV、Machine Learning、Medical Image Analysis、Bioinformatics、ICML、NIPS、SIGKDD、IJCAI、AAAI等）上发表论文百余篇，Google Scholar总引用3000余次。常年担任相关领域顶级期刊和会议的审稿人或程序委员，现担任Information Science等多个SCI期刊的编委。

  　王立威 北京大学

报告题目：The Margin Theory of Machine Learning

报告内容： 在Vapnik等人创立的经典margin理论中，SVM（或任何线性分类器）的性能只与margin有关，而与数据所在的空间维数无关。这一理论使得我们可以利用核方法将输入空间映射到（无穷维）特征空间进行线性分类。在本报告中，我将证明这一机器学习经典结论并不完全正确，SVM性能并非仅由margin决定而与特征空间维数无关。具体的，我将证明一个基于与特征空间维数相关的margin上界。该上界一致紧于经典的维数无关margin上界；当特征空间维数是无穷大时，新上界等价于传统维数无关margin上界。这一margin理论表明，核方法为了提高margin而增加特征空间维数时，一定程度上付出了性能的代价。实验结果显示该理论对于SVM核函数的选择具有指导意义。

讲者简介： 王立威，北京大学信息科学技术学院教授。于清华大学电子工程系获本科和硕士学位，北京大学数学学院获博士学位。自2005年起在北京大学信息学院任教。主要研究兴趣为机器学习理论。在机器学习重要会议COLT, NIPS, ICML和期刊JMLR, IEEE Trans. PAMI等发表论文40余篇。2008年发表于COLT的论文On the Margin Explanation of Boosting Algorithms是中国大陆学者在该会议上的首篇论文。2010年入选AI’s 10 to Watch，是首位获此荣誉的亚洲学者。2012年获得首届国家自然科学基金优秀青年基金。目前任中国计算机学会模式识别与人工智能专委会委员，中国人工智能学会模式识别专委会委员。担任Journal of Computer Science and Technology (JCST)等期刊编委。

   　　王瑞平 中科院计算所

报告题目：Learning on Riemannian Manifold for Video-based Face Recognition

报告内容： Recently the ubiquitous use of video capturing devices is shifting the focus of face recognition research from image-based scenarios to video-based ones. In this talk, I will introduce recent progresses in our group towards this topic. By simply treating video as image set, our works mainly deal with the problem of robust image set modeling and efficient metric learning for classification. Specifically, for robust set modeling, we propose to use the natural second-order statistic - covariance matrix (a.k.a SPD matrix) and further the Gaussian distribution model as set features to characterize the data structure of the image set. Then under the framework of Riemannian geometry and information geometry, the covariance matrices or Gaussian models are embedded into some certain Riemannian manifolds, where Riemannian kernels can be derived based on valid Riemannian metrics and thus facilitate learning algorithms originally developed in Euclidean space. For set classification, the problem is naturally formulated as discriminative metric learning on such Riemannian manifolds. Here, the notion “metric learning” is associated with different application scenarios. We have considered both cases of video vs. video matching and still image vs. video (e.g. image as gallery and video as probe). For large scale face video retrieval applications, we further develop corresponding hashing methods to learn discriminative compact binary codes for highly efficient video search. Representative works appear at IEEE CVPR2008/2009/2012/2014/2015.

讲者简介： Ruiping Wang is an Associate Professor at the Institute of Computing Technology (ICT), Chinese Academy of Sciences (CAS). Prior to joining ICT in July 2012, he was a postdoctoral researcher with the Tsinghua University from July 2010 to June 2012. He also spent one year working as a Research Associate with the University of Maryland, College Park, from Nov. 2010 to Oct. 2011. He received the B.S. degree in Applied Mathematics from Beijing Jiaotong University in 2003, and the Ph.D. degree in Computer Science from ICT, CAS, in 2010. He has published more than 30 papers in peer-reviewed journals and conferences, including IEEE TPAMI, TIP, TMM, PR, CVPR, ICCV, and has received the Best Student Poster Award Runner-up from IEEE CVPR 2008 for the work on Manifold-Manifold Distance. He serves as regular reviewer/PC member for a number of leading journals and conferences, e.g. IEEE TPAMI, TIP, TCSVT, TMM, TNNLS, IJCV, ICCV, CVPR, ECCV. He has organized tutorials in ACCV 2014 and CVPR 2015 with his colleagues. He has given invited talks in workshops of ICME 2014 and ACCV 2014. His current research interests include video-based face recognition, facial expression analysis, image set classification, distance metric learning, and manifold learning.

   　　王甦菁 中科院心理所

报告题目：An Introduction of Micro-expression Recognition

报告内容： 在这个报告中，我将给大家简单介绍一下微表情识别的一些进展。微表情指人们在试图隐藏自己的情绪时所泄露快速表情。这样微表情就可以作为一条很 重要的测谎线索用于测谎，并有可能被广泛地应用于安全、司法临床等领域。在本报告中，列举了当前微表情识别几个公开的数据库，并介绍了几种微表情识别的新的算法，最后了微表情识别中所面临的问题。

讲者简介： 王甦菁，2012 年 7 月博士毕业于吉林大学计算机科学与技术学院，同年 8月进入中国科学院心理研究所从事博士后研究，长期从事特征抽取在人脸识别上应用基础研究，在流形学习，张量分析，彩色空间理论等方面有着深入的研究，其成果发表在本领域国际权威期刊IEEE Transactions 系列期刊上发表多篇论文。担任国际学术期刊Neurocomputing 的编委。承担 1 项国家自然科学基金面上项目“基于稀疏张量的微表情识别研究”项目，发表 30 余篇论文。

   　　杨易 悉尼科技大学

报告题目：大规模视频数据复杂事件检测

报告内容： Compared to visual concepts such as actions, scenes and objects, complex event is a higher level abstraction of longer video sequences. For example, a “marriage proposal” event is described by multiple objects (e.g., ring, faces), scenes (e.g., in a restaurant, outdoor) and actions (e.g., kneeling down). I will present our work on detecting more complicated and generic events that gain more users’ interest. The positive exemplars which exactly convey the precise semantic of an event are hard to obtain. It would be beneficial to utilize the related exemplars for complex event detection. However, the semantic correlations between related exemplars and the target event vary substantially as relatedness assessment is subjective. Two related exemplars can be about completely different events, e.g., in the TRECVID MED dataset, both bicycle riding and equestrianism are labelled as related to “attempting a bike trick” event. To tackle the subjectiveness of human assessment, our algorithm automatically evaluates how positive the related exemplars are for the detection of an event and uses them on an exemplar-specific basis. Experiments demonstrate that our algorithm is able to utilize related exemplars adaptively, and the algorithm gains good performance for complex event detection. As the information from a few positive examples is limited, we propose using knowledge adaptation to facilitate event detection. Different from the state of the art, our algorithm is able to adapt knowledge from another source for MED even if the features of the source and the target are partially different, but overlapping. Avoiding the requirement that the two domains are consistent in feature types is desirable as data collection platforms change or augment their capabilities and we should be able to respond to this with little or no effort. We also makes two contributions to the inference of CNN video representation. First, while average pooling and max pooling have long been the standard approaches to aggregating frame level static features, we show that performance can be significantly improved by taking advantage of an appropriate encoding method. Second, we propose using a set of latent concept descriptors as the frame descriptor, which enriches visual information while keeping it computationally affordable.

讲者简介： 杨易现为悉尼科技大学高级讲师。2010年在浙江大学获得博士学位，导师为潘云鹤院士和庄越挺教授。2011年至2013年在卡内基梅隆大学计算机学院做博士后研究员。2012年获得教育部全国优秀博士论文奖。2013年获得Australia Research Council 颁发的Discover Early Career Researcher Award。主要研究方向为大规模视频分析与理解、跨媒体智能计算、监控视频分析和智能看护技术。

   　　张道强 南京航空航天大学

报告题目：脑影像／脑网络智能分析方法及应用

报告内容：近年来，“脑科学计划”吸引了各国政府和公众的广泛关注。脑影像技术是研究脑科学的重要工具之一, 然而由于脑影像数据所固有的高维度、多模态、异构和时变等特性，对其进行快速有效分析是当前研究的关键问题之一。在本报告中，我们将首先简要介绍脑影像／脑网络分析的基本方法，然后重点介绍我们近几年在基于机器学习的脑影像／脑网络智能分析方面的相关工作，并介绍其在老年痴呆症等脑疾病的早期诊断中的应用。

讲者简介：张道强，工学博士，南京航空航天大学教授，博士生导师。分别于1999年和2004年在南京航空航天大学计算机科学与工程系获学士和博士学位，2006年在南京大学计算机软件新技术国家重点实验室博士后出站。2004年起在南京航空航天大学任教，2008年破格晋升为教授，主要研究方向为机器学习和模式识别技术及应用。2010年至2012年在美国北卡罗莱纳大学教堂山分校（UNC-Chapel Hill）从事脑影像分析及脑疾病早期诊断研究。先后主持多项国家和省部级基金，已在国内外核心期刊和会议上发表100余篇论文，论文累计被他引4000余次（Google Scholar），研究成果获得多次国际奖项，包括国际期刊《Pattern Recognition》 2006-2010年高引用论文奖、国际会议PRICAI'06及STMI'12最佳论文奖等。目前担任《PLOS ONE》等期刊编委，多家学术期刊的审稿专家及多个国际会议的程序委员会委员。任中国人工智能学会机器学习专委会常委、中国计算机学会人工智能与模式识别专委会委员等职务。曾获2006年全国优秀博士学位论文提名奖，2012年霍英东基金会第十三届高等院校青年教师奖，2013年江苏省杰出青年基金及2014年国家自然科学基金优秀青年基金。2014年入选Elsevier中国高被引学者榜单（计算机科学）。

   　　张开华 南京信息工程大学

报告题目：Designing Fast Appearance Model for Visual Tracking

报告内容： Since the target appearance will change significantly due to complex factors in the scene like illumination change, pose variation, occlusion, etc., an effective appearance model plays a key role in visual tracking. On the other hand, a fast appearance model is a prerequisite for real-time tracking, which can be readily applied to some real-time applications, such as visual surveillance, human-computer interface, etc. In this talk, I will briefly introduce some of our recently developed methods that exploit the efficient filtering approach to design fast appearance models, including those without learning based on compressive sensing theory and convolutional networks, and that with very simple spatio-temporal context learning method.

讲者简介： 博士，教授。2006年于中国海洋大学电子工程系获学士学位；2009年于中国科学技术大学电子工程与信息科学系获硕士学位；2014在香港理工大学计算机系获博士学位。其中，2009年8月到2010年8月在香港理工大学计算机系从事研究工作，任研究助理。2013年1月至今，工作于南京信息工程大学信息与控制学院。主要研究领域包括水平集图像分割、视觉目标跟踪等。已在在国内外学术期刊和会议上发表论文20余篇，其中包括：IEEE T-PAMI/T-IP/T-CSVT/T-Cyb/T-GRS等国际期刊论文。据Google Scholar统计，论文近五年被引用1260余次。