学术讲座通知​:From Shuffled Linear Regression to Homomorphic Sensing

题目:From Shuffled Linear Regression to Homomorphic Sensing
报告人:Dr. Manolis Tsakiris, 上海科技大学
时间:2019年5月30日14:00-15:30 (星期四)
地点:教三 308  主持人:李春光

A recent line of research termed Shuffled Linear Regression has been exploring under great generality the recovery of signals from permuted measurements; a challenging problem in diverse fields of data science and machine learning. In its simplest form it consists of solving a linear system of equations for which the right-hand-side vector has been permuted. In the first part of this talk I will present a provably correct method based on algebraic geometry together with its associated algorithm, the latter being a first working solution to this open problem, able to handle thousands of noisy fully permuted measurements in milliseconds. In the second part of the talk I will discuss the issue of uniqueness of the solution, in a general context which I have termed Homomorphic Sensing*. Given a linear subspace and a finite set of linear transformations I will present dimension conditions of algebraic-geometric nature guaranteeing that points in the subspace are uniquely determined from their homomorphic image under some transformation in the set. As a special case, this theory explains the operational regime of Unlabeled Sensing, in which the goal is unique recovery of signals from both permuted and subsampled measurements.
*Has been accepted by ICML2019. Preprint: https://arxiv.org/abs/1901.07852

Manolis Tsakiris is an electrical engineering and computer science graduate of the National Technical University of Athens, Greece. He holds an M.S. degree in signal processing from Imperial College London, UK, and a Ph.D. degree from Johns Hopkins University, USA, in theoretical machine learning, under the supervision of Prof. Rene Vidal. Since August 2017 he is an assistant professor at the School of Information Science and Technology (SIST) at ShanghaiTech University. His main research interests are subspace learning methods and related problems in algebraic geometry. For more information, please visit his homepage.



题目:Modeling deep structures with application to object detection and pose estimation
报告人:欧阳万里 博士 香港中文大学

摘要:Deep learning attempts to learn feature representation by multiple levels of abstraction. It is found to be useful in speech recognition, face recognition, image classification, biology, physics, and material science. In this talk, a brief introduction will be given on our recent progress in using deep learning as a tool for modeling the structure in visual data for object detection and human pose estimation. We show that observation in our problem are useful in modeling the structure of deep model and help to improve the performance of deep models for our problem.

Wanli Ouyang received the PhD degree in the Department of Electronic Engineering, The Chinese University of Hong Kong, where he is now a research assistant professor. His research interests include image processing, computer vision and pattern recognition. He is the first/correspondence author of 6 papers on TPAMI and IJCV, and has published 26 papers on top tier conferences like CVPR, ICCV and NIPS. ImageNet Large Scale Visual Recognition Challenge (ILSVRC) is one of the most important grand challenges in computer vision. The team led by him ranks No. 1 in the ILSVRC 2015 and ILSVRC 2016. He receives the best reviewer award of ICCV. He has been the reviewer of many top journals and conferences such as IEEE TPAMI, TIP, IJCV, TSP, TITS, TNN, CVPR, and ICCV. He is a senior member of the IEEE.
(更多信息请关注: http://www.ee.cuhk.edu.hk/~wlouyang/ )


报告人凌海滨 博士(美国天普大学副教授 亮风台科技的联合创始人&首席科学家)

报告人简介: 凌海滨博士于1997年和2000年毕业于北京大学,于2006年在美国马里兰大学(Maryland University)获博士学位,后于2006~2007年在加州大学洛杉矶分校(UCLA)做博士后。在2001年任微软亚洲研究院(MSRA)助理研究员,2007~2008年任西门子研究院研究员,从2008起任职于美国天普大学(Temple University),现在为计算机系副教授,并且是亮风台科技的联合创始人并担任其首席科学家。主要研究领域为计算机视觉、增强现实、医学图像理解、和人机交互。获2003年度ACM UIST最佳学生论文奖,2014年度美国自然科学基金CAREER Award。担任期刊IEEE Trans. on Pattern Analysis and Machine Intelligence和Pattern Recognition的编委,以及CVPR 2014和CVPR 2016年的领域主席。
更多信息请参阅: http://www.dabi.temple.edu/~hbling/

学术报告: Complete Dictionary Recovery over the Sphere

报告题目Complete Dictionary Recovery over the Sphere
报告人:Dr.Ju Sun, Electrical Engineering, Columbia University
主持人: 北邮模式识别实验室 李春光
时间:2015年9月2日 15:00-16:30

报告摘要:We consider the problem of recovering a complete (i.e., square and invertible) matrix $A_0$, from $Y \in R^{n\times p} with $Y = A_0X_0$, provided $X_0$ is sufficiently sparse. This recovery problem is central to the theoretical understanding of dictionary learning, which seeks a sparse representation for a collection of input signals, and finds numerous applications in modern signal processing and machine learning. We give the first efficient algorithm that provably recovers $A_0$ when $X_0$ has $O (n)$ nonzeros per column, under suitable probability model for $X_0$. In contrast, prior results based on efficient algorithms provide recovery guarantees when $X_0$ has only $O (pn)$ nonzeros per column. Our algorithmic pipeline centers around solving a certain nonconvex optimization problem with a spherical constraint, and hence is naturally phrased in the language of manifold optimization. To show this apparently hard problem is tractable, we first provide a geometric characterization of the high-dimensional objective landscape, which shows that with high probability there are no “spurious” local minima. This particular geometric structure allows us to design a Riemannian trust region algorithm over the sphere that provably converges to one local minimizer with an arbitrary initialization, despite the presence of saddle points. The geometric approach we develop here may also shed light on other problems arising from nonconvex recovery of structured signals.

报告人简介:Ju Sun is now a five year PhD candidate (Advisor: Prof. John Wright) in the Department of Electrical Engineering, Columbia University in the City of New York. He works at the intersection of computer vision, machine learning, numerical optimization, signal/image processing, information theory, and compressive sensing, focusing on modeling, harnessing, and computing with structures in massive data, with provable guarantees and practical algorithms. His paper “Complete Dictionary Recovery over the Sphere” has recently received the best student paper award from SPARS 2015, which was held in Cambridge University.



学术报告: Subspace Clustering – Recent Advances

报告题目:Subspace Clustering – Recent Advances
报告人:Zhouchen Lin(林宙辰)博士 北京大学教授
地点:教三 810会议室

报告摘要: Nowadays we are in the big data era, where the data is usually high dimensional. How to process high dimensional data effectively is a critical issue. Fortunately, we observe that data usually distribute near low dimensional manifolds. Mixture of subspaces is a simple yet effective model to represent high dimensional data, where the membership of the data points to the subspaces might be unknown. Therefore, there is a need to simultaneously cluster the data into multiple subspaces and find a low-dimensional subspace fitting each group of data points. This problem, known as subspace clustering, has found numerous applications. In this talk, I will present my recent work on this research problem.

ZHOUCHEN LIN received the Ph.D. degree in applied mathematics from Peking University in 2000. He is currently a Professor at Key Laboratory of Machine Perception (MOE), School of Electronics Engineering and Computer Science, Peking University. He is also a Chair Professor at Northeast Normal University and a guest professor at Beijing Jiaotong University. Before March 2012, he was a Lead Researcher at Visual Computing Group, Microsoft Research Asia. He was a guest professor at Shanghai Jiaotong University and Southeast University, and a guest researcher at Institute of Computing Technology, Chinese Academy of Sciences. His research interests include computer vision, image processing, computer graphics, machine learning, pattern recognition, and numerical computation and optimization. He is an associate editor of International J. Computer Vision and a Senior member of the IEEE. He served CVPR2014 as an area chair.

学术报告通知: Bayesian Inference with Max-margin Posterior Regularization

报告题目: Bayesian Inference with Max-margin Posterior Regularization
报告人:朱军 副教授 清华大学智能技术与系统国家重点实验室智能媒体组
时间:2012年11月23日 10:00-11:30
报告摘要:Existing Bayesian models, especially nonparametric Bayesian methods, rely heavily on specially conceived priors to incorporate domain knowledge for discovering improved latent representations. While priors can affect posterior distributions through Bayes’ theorem, imposing posterior regularization is arguably more direct and in some cases can be more natural and easier. In this talk, I will present regularized Bayesian inference (RegBayes), a computational framework to perform posterior inference with a convex regularization on the desired post-data posterior distributions. When the convex regularization is induced from a linear operator on the posterior distributions, RegBayes can be solved with convex analysis theory. Furthermore, I will present some concrete examples, including MedLDA for learning discriminative topic representations and infinite latent support vector machines for learning discriminative latent features for classification. All these models explore the large-margin idea in combination with a (nonparametric) Bayesian model for discovering predictive latent representations. I will discuss both variational and Monte Carlo methods for approximate inference.
报告人简介:Dr. Jun Zhu is an associate professor in the Department of Computer Science and Technology at Tsinghua University. His principal research interests lie in the development of statistical machine learning methods for solving scientific and engineering problems arising from artificial and biological learning, reasoning, and decision-making in the high-dimensional and dynamic worlds. Prof. Zhu received his Ph.D. in Computer Science from Tsinghua University, and his advisor was Prof. Bo Zhang. He did post-doctoral research with Prof. Eric P. Xing in the Machine Learning Department at Carnegie Mellon University. His current work involves both the foundations of statistical learning, including theory and algorithms for probabilistic latent variable models, sparse learning in high dimensions, Bayesian nonparametrics, and large-margin learning; and the application of statistical learning in social network analysis, data mining, and multi-media data analysis.

学术报告通知:Rank Minimization: A Way to 2D Sparsity



报告题目: Rank Minimization: A Way to 2D Sparsity

报告人:林宙辰博士  微软亚洲研究院视觉计算组主管研究员(Lead Researcher)

时间:2010年12月22日 15:00-16:30

地点:教三楼 411


报告摘要:Sparse representation has been a hot topic in signal processing and machine learning in recent years. In the past, people usually discussed 1D sparsity, i.e., the number of nonzeros in a vector. For 2D signals, we can actually also define a measure of sparsity, namely the rank of the data matrix. In this talk, I will give a brief history of sparse representation and then introduce some core problems and theories of rank minimization. Finally, I will show some applications of rank minimization that have resulted in improved performance in handling imperfect data that can have noise, outliers and missing values.


报告人简介:Dr. Zhouchen Lin is a Lead Researcher at Visual Computing Group, Microsoft Research Asia. He received the Ph.D. degree in applied mathematics from PekingUniversity in 2000. He is now a guest professor to ShanghaiJiaotongUniversity, BeijingJiaotongUniversity and SoutheastUniversity. He is also a guest researcher to Institute of Computing Technology, ChineseAcademy of Sciences. His research interests include computer vision, image processing, computer graphics, machine learning, pattern recognition, and numerical computation and optimization. He is a senior member of the IEEE.