- talks - lists.cs.princeton.edu

Colloquium Speaker: Leonidas Guibas, Today, 12:30pm- reminder
by Nicole E. Wagenblast 25 Feb '16

25 Feb '16

Colloquium Speaker Leonidas Guibas, from Stanford University Thursday, February 25, 2016 - 12:30pm Computer Science 105 Networks of Shapes and Images Multimedia content has become a ubiquitous presence on all our computing devices, spanning the gamut from live content captured by device sensors such as smartphone cameras to immense databases of images, audio, video, 3D scans and 3D models stored in the cloud. As we try to maximize the utility and value of all these petabytes of content, we often do so by analyzing each piece of data individually and foregoing a deeper analysis of the relationships between the media. In this talk we focus on developing rigorous mathematical and computational tools for making such relationships or correspondences between signal and media data sets first-class citizens -- so that the relationships themselves become explicit, algebraic, storable and searchable objects. We discuss mathematical and algorithmic issues on how to represent and compute relationships or mappings at multiple levels of detail -- and go on to build entire networks based on these relationships in collections of inter-related data. Information transport and aggregation in such networks naturally lead to abstractions of objects and other visual entities, allowing data compression while capturing variability as well as shared structure. Furthermore, the network can act as a regularizer, allowing us to to benefit from the "wisdom of the collection" in performing operations on individual data sets or in map inference between them, ultimately enabling a certain joint understanding of data that provides the powers of abstraction, analogy, compression, error correction, and summarization. Examples include entity extraction from images or videos, 3D segmentation, the propagation of annotations and labels among images/videos/3D models, variability analysis in a collection of shapes, etc. Finally we briefly describe the ShapeNet effort, an attempt to build a large-scale repository of 3D models richly annotated with geometric, physical, functional, and semantic information -- both individually and in relation to other models and media. More than a repository, ShapeNet is a true network that allows information transport not only between its nodes but also to/from new visual data coming from sensors. This effectively enables us to add missing information to signals, giving us for example the ability to infer what an occluded part of an object in an image may look like, or what other object arrangements may be possible, based on the world-knowledge encoded in ShapeNet. This is joint work with several collaborators, as will be discussed during the talk. Leonidas Guibas obtained his Ph.D. from Stanford under the supervision of Donald Knuth. His main subsequent employers were Xerox PARC, DEC/SRC, MIT, and Stanford. He is currently the Paul Pigott Professor of Computer Science (and by courtesy, Electrical Engineering) at Stanford University. He heads the Geometric Computation group and is part of the Graphics Laboratory, the AI Laboratory, the Bio-X Program, and the Institute for Computational and Mathematical Engineering. Professor Guibas’ interests span geometric data analysis, computational geometry, geometric modeling, computer graphics, computer vision, robotics, ad hoc communication and sensor networks, and discrete algorithms. Some well-known past accomplishments include the analysis of double hashing, red-black trees, the quad-edge data structure, Voronoi-Delaunay algorithms, the Earth Mover’s distance, Kinetic Data Structures (KDS), Metropolis light transport, heat-kernel signatures, and functional maps. Professor Guibas is an ACM Fellow, an IEEE Fellow and winner of the ACM Allen Newell award. _______________________________________________ talks mailing list talks(a)lists.cs.princeton.edu To edit subscription settings or remove yourself, use this link: https://lists.cs.princeton.edu/mailman/listinfo/talks

1 0

Colloquium Speaker: Leonidas Guibas, Thursday, Feburary 25, 12:30pm- Reminder
by Nicole E. Wagenblast 24 Feb '16

24 Feb '16

Colloquium Speaker Leonidas Guibas, from Stanford University Thursday, February 25, 2016 - 12:30pm Computer Science 105 Networks of Shapes and Images Multimedia content has become a ubiquitous presence on all our computing devices, spanning the gamut from live content captured by device sensors such as smartphone cameras to immense databases of images, audio, video, 3D scans and 3D models stored in the cloud. As we try to maximize the utility and value of all these petabytes of content, we often do so by analyzing each piece of data individually and foregoing a deeper analysis of the relationships between the media. In this talk we focus on developing rigorous mathematical and computational tools for making such relationships or correspondences between signal and media data sets first-class citizens -- so that the relationships themselves become explicit, algebraic, storable and searchable objects. We discuss mathematical and algorithmic issues on how to represent and compute relationships or mappings at multiple levels of detail -- and go on to build entire networks based on these relationships in collections of inter-related data. Information transport and aggregation in such networks naturally lead to abstractions of objects and other visual entities, allowing data compression while capturing variability as well as shared structure. Furthermore, the network can act as a regularizer, allowing us to to benefit from the "wisdom of the collection" in performing operations on individual data sets or in map inference between them, ultimately enabling a certain joint understanding of data that provides the powers of abstraction, analogy, compression, error correction, and summarization. Examples include entity extraction from images or videos, 3D segmentation, the propagation of annotations and labels among images/videos/3D models, variability analysis in a collection of shapes, etc. Finally we briefly describe the ShapeNet effort, an attempt to build a large-scale repository of 3D models richly annotated with geometric, physical, functional, and semantic information -- both individually and in relation to other models and media. More than a repository, ShapeNet is a true network that allows information transport not only between its nodes but also to/from new visual data coming from sensors. This effectively enables us to add missing information to signals, giving us for example the ability to infer what an occluded part of an object in an image may look like, or what other object arrangements may be possible, based on the world-knowledge encoded in ShapeNet. This is joint work with several collaborators, as will be discussed during the talk. Leonidas Guibas obtained his Ph.D. from Stanford under the supervision of Donald Knuth. His main subsequent employers were Xerox PARC, DEC/SRC, MIT, and Stanford. He is currently the Paul Pigott Professor of Computer Science (and by courtesy, Electrical Engineering) at Stanford University. He heads the Geometric Computation group and is part of the Graphics Laboratory, the AI Laboratory, the Bio-X Program, and the Institute for Computational and Mathematical Engineering. Professor Guibas’ interests span geometric data analysis, computational geometry, geometric modeling, computer graphics, computer vision, robotics, ad hoc communication and sensor networks, and discrete algorithms. Some well-known past accomplishments include the analysis of double hashing, red-black trees, the quad-edge data structure, Voronoi-Delaunay algorithms, the Earth Mover’s distance, Kinetic Data Structures (KDS), Metropolis light transport, heat-kernel signatures, and functional maps. Professor Guibas is an ACM Fellow, an IEEE Fellow and winner of the ACM Allen Newell award.

1 0

Colloquium-Greg Durrett, Tuesday, March 1, 12:30PM
by Mitra D. Kelly 23 Feb '16

23 Feb '16

Colloquium Speaker Greg Durrett, UC Berkeley Tuesday, March 1st, 12:30PM Cs 105 Data-Driven Text Analysis with Joint Models One reason that analyzing text is hard is that it involves dealing with deeply entangled linguistic variables: objects like syntactic structures, semantic types, and discourse relations depend on one another in complex ways. Our work tackles several facets of text analysis using joint modeling, combining model components both across and within the various subtasks of this analysis. This model structure allows us to pass information between these entangled subtasks and propagate high-confidence predictions rather than errors. Critically, our models have the capacity to learn key linguistic phenomena as well as other important patterns in the data; that is, linguistics tells us how to structure these models, then the data injects knowledge into them. We describe state-of-the-art systems for a range of tasks, including syntactic parsing, entity resolution, and document summarization. Bio: Greg is a Ph.D. candidate at UC Berkeley working on natural language processing with Dan Klein. He is interested in building structured machine learning models for a wide variety of text analysis problems and downstream NLP applications. His work is comprised of two broad thrusts: first, designing joint models that combine information across different tasks or different views of a problem, and second, building systems that strike a balance between being linguistically motivated and data-driven Mitra Kelly Academic Secretary Princeton University Computer Science Dept 35 Olden Street Princeton NJ 08540 mkelly(a)cs.princeton.edu <mailto:mkelly@cs.princeton.edu> 609-258-4562

1 0

Colloquium Speaker Nikolay Atanasov, Friday Feb 26, 3:30pm
by Nicole E. Wagenblast 23 Feb '16

23 Feb '16

Colloquium Speaker Dr. Nikolay Atanasov Acquiring Metric and Semantic Information using Autonomous Robots Friday, February 26th at 3:30 PM Maeder Hall, ACEE Recent years have seen impressive progress in robot control and perception including adept manipulation, aggressive quadrotor maneuvers, dense metric map reconstruction, and object recognition in real time. The grand challenge in robotics today is to capitalize on these advances in order to enable autonomy at a higher-level of intelligence. It is compelling to envision teams of autonomous robots in environmental monitoring, precision agriculture, construction and structure inspection, security and surveillance, and search and rescue. In this talk, I will emphasize that many such applications can be addressed by thinking about how to coordinate robots in order to extract useful information about the environment. More precisely, I will formulate a general active estimation problem that captures the common characteristics of the aforementioned scenarios. I will show how to manage the complexity of the problem over metric information spaces with respect to long planning horizons and large robot teams. These results lead to computationally scalable, non-myopic algorithms with quantified performance for problems such as distributed source seeking and active simultaneous localization and mapping (SLAM). I will then focus on acquiring information using both metric and semantic observations (e.g., object recognition). In this context, there are several new challenges such as missed detections, false alarms, and unknown data association. To address them, I will model semantic observations via random sets and will discuss filtering using such models. A major contribution of our approach is in proving that the complexity of the problem is equivalent to computing the permanent of a suitable matrix. This enables us to develop and experimentally validate algorithms for semantic localization, mapping, and planning on mobile robots, Google's project Tango phone, and the KITTI visual odometry dataset.

1 0

Colloquium speaker Yuxin Chen, TODAY, 4:30pm
by Nicole E. Wagenblast 23 Feb '16

23 Feb '16

Colloquium Speaker Yuxin Chen, Stanford University Monday, February 29, 4:30pm Computer Science 104- large auditorium The Power of Nonconvex Paradigms for High-Dimensional Estimation In various scenarios in the information sciences, one wishes to estimate a large number of parameters from highly incomplete / imperfect data samples. A growing body of recent work has demonstrated the effectiveness of convex relaxation --- in particular, semidefinite programming --- for solving many problems of this kind. However, the computational cost of such convex paradigms is often unsatisfactory, which limits applicability to large-dimensional data. This talk follows another route: by formulating the problems into nonconvex programs, we attempt to optimize the nonconvex objectives directly. To illustrate the power of this strategy, we present two concrete stories. The first involves solving a random quadratic system of equations, which spans many applications ranging from the century-old phase retrieval problem to various latent-variable models in machine learning. The second is about recovering a collection of discrete variables from noisy pairwise difference measurements, which arises when one wishes to jointly align multiple images or to retrieve the genome phases from paired sequencing reads. We propose novel nonconvex paradigms for solving these two problems. In both cases, the proposed solutions can be accomplished within linear time, while achieving a statistical accuracy that is nearly un-improvable. Yuxin Chen is currently a postdoctoral scholar in the Department of Statistics at Stanford University, supervised by Prof. Emmanuel Candès. He received the Ph.D. degree in Electrical Engineering and M.S. in Statistics from Stanford University, M.S. in Electrical and Computer Engineering from the University of Texas at Austin, and B.E. in microelectronics from Tsinghua University. His research interests include high-dimensional structured estimation, convex and nonconvex optimization, statistical learning, and information theory.

1 0

Colloquium speaker Yuxin Chen, Monday Feb 29, 4:30pm
by Nicole E. Wagenblast 23 Feb '16

23 Feb '16

Colloquium Speaker Yuxin Chen, Stanford University Monday, February 29, 4:30pm Computer Science 104- large auditorium The Power of Nonconvex Paradigms for High-Dimensional Estimation In various scenarios in the information sciences, one wishes to estimate a large number of parameters from highly incomplete / imperfect data samples. A growing body of recent work has demonstrated the effectiveness of convex relaxation --- in particular, semidefinite programming --- for solving many problems of this kind. However, the computational cost of such convex paradigms is often unsatisfactory, which limits applicability to large-dimensional data. This talk follows another route: by formulating the problems into nonconvex programs, we attempt to optimize the nonconvex objectives directly. To illustrate the power of this strategy, we present two concrete stories. The first involves solving a random quadratic system of equations, which spans many applications ranging from the century-old phase retrieval problem to various latent-variable models in machine learning. The second is about recovering a collection of discrete variables from noisy pairwise difference measurements, which arises when one wishes to jointly align multiple images or to retrieve the genome phases from paired sequencing reads. We propose novel nonconvex paradigms for solving these two problems. In both cases, the proposed solutions can be accomplished within linear time, while achieving a statistical accuracy that is nearly un-improvable. Yuxin Chen is currently a postdoctoral scholar in the Department of Statistics at Stanford University, supervised by Prof. Emmanuel Candès. He received the Ph.D. degree in Electrical Engineering and M.S. in Statistics from Stanford University, M.S. in Electrical and Computer Engineering from the University of Texas at Austin, and B.E. in microelectronics from Tsinghua University. His research interests include high-dimensional structured estimation, convex and nonconvex optimization, statistical learning, and information theory.

1 0

colloquium, Alp Kucukelbir , today Feb 23rd at 12:30pm
by Mitra D. Kelly 23 Feb '16

23 Feb '16

Colloquium Alp Kucukelbir, Columbia Feb 23rd @12:30pm Cs 105 # Talk Title Towards Automated Machine Learning # Talk Abstract Machine learning is changing the way we do science. We want to study large datasets to shed light onto natural processes. (How do proteins function? How does the brain work?) To this end, we need tools to rapidly and iteratively explore hidden patterns in data. However, using machine learning to discover such patterns requires enormous effort and cross-disciplinary expertise. My goal is to develop easy-to-use machine learning tools that empower scientists to gain new insights from data. This requires research in automated algorithms, scalable software, and robust machine learning. These are the building blocks of effective machine learning tools. # Bio Alp is a postdoctoral research scientist at the Data Science Institute and the department of Computer Science at Columbia University. He works with David Blei on developing scalable and robust machine learning tools. He collaborates with Andrew Gelman on the Stan probabilistic programming system. Alp received his Ph.D. from Yale University, where he was awarded the Becton prize for best thesis in engineering and applied science. He holds a B.A.Sc. from the University of Toronto. Mitra Kelly Academic Secretary Princeton University Computer Science Dept 35 Olden Street Princeton NJ 08540 mkelly(a)cs.princeton.edu <mailto:mkelly@cs.princeton.edu> 609-258-4562

1 0

Colloquium Speaker -Alp Kucukelbir, Columbia University, Tuesday, Feb 23
by Mitra D. Kelly 22 Feb '16

22 Feb '16

Colloquium Speaker Alp Kucukelbir, Columbia University Tuesday, Feb 23 at 12:30 CS 105 # Talk Title Towards Automated Machine Learning # Talk Abstract Machine learning is changing the way we do science. We want to study large datasets to shed light onto natural processes. (How do proteins function? How does the brain work?) To this end, we need tools to rapidly and iteratively explore hidden patterns in data. However, using machine learning to discover such patterns requires enormous effort and cross-disciplinary expertise. My goal is to develop easy-to-use machine learning tools that empower scientists to gain new insights from data. This requires research in automated algorithms, scalable software, and robust machine learning. These are the building blocks of effective machine learning tools. # Bio Alp is a postdoctoral research scientist at the Data Science Institute and the department of Computer Science at Columbia University. He works with David Blei on developing scalable and robust machine learning tools. He collaborates with Andrew Gelman on the Stan probabilistic programming system. Alp received his Ph.D. from Yale University, where he was awarded the Becton prize for best thesis in engineering and applied science. He holds a B.A.Sc. from the University of Toronto. Mitra Kelly Academic Secretary Princeton University Computer Science Dept 35 Olden Street Princeton NJ 08540 mkelly(a)cs.princeton.edu <mailto:mkelly@cs.princeton.edu> 609-258-4562

1 0

4:30pm Thu Feb 25 talk on "information theory at the heart of data"
by Jennifer Rexford 22 Feb '16

22 Feb '16

> > > Speaker: > Hamed Hassani, ETH Zurich > Title: > From Communication to Sensing and Learning: Information Theory at the Heart of Data Science > > Day: > Thursday, February 25, 2016 > > > Time: > 4:30 pm > Room: > B205, E-Quad > Host: > Prof. Mung Chiang > > > > > > > > > > Abstract: > We are witnessing a new era of science — ushered in by our ability to collect massive amounts of data and unprecedented ways to learn about the physical world. Beyond the challenges of storage and communication, there are new frontiers in the acquisition, analysis and exploration of the data. In this talk, I will view these frontiers through the lens of information theory. I will argue that information theory lies at the center of data science, offering insights beyond its classical applications. As a concrete example, I will consider the problem of optimal data acquisition, a challenge which arises in active learning, optimal sensing and experimental design. Based on information theoretic foundations, and equipped with tools from submodular optimization theory, I will present a rigorous analysis of the widely-used sequential information maximization policy (also known as the information-gain heuristic). Our analysis establishes conditions under which this policy provably works near-optimally and identifies situations where the policy fails. In the latter case, our framework suggests novel, efficient surrogate objectives and algorithms that outperform classical techniques. > > Bio: > Hamed Hassani is a post-doctoral scholar at the Institute for Machine Learning at ETH Zurich. He received a Ph.D. degree in Computer and Communication Sciences from EPFL, Lausanne. Prior to that, he received a B.Sc. degree in Electrical Engineering and a B.Sc. degree in Mathematics from Sharif University of Technology, Tehran. Hamed's fields of interest include machine learning, coding and information theory as well as theory and applications of graphical models. He is the recipient of the 2014 IEEE Information Theory Society Thomas M. Cover Dissertation Award. His co-authored paper at NIPS 2015 was selected for an oral (plenary) presentation, and his co-authored paper at ISIT 2015 received the IEEE Jack Keil Wolf ISIT Student Paper Award. > > > > > Heather Evans > Assistant to the Chair > Department of Electrical Engineering > Princeton University > Engineering Quadrangle, B-210 > Olden Street > Princeton, NJ 08544 > (p): 609-258-7282 > (f): 609-258-0119 > heathere(a)princeton.edu <mailto:heathere@princeton.edu> > > >

1 0

Young Lee will present his FPO, "Targeted analyses of very large genome-wide data collections" on Tuesday, 3/1/2016 at 10:30am in CS 402
by Nicki Gotsis 22 Feb '16

22 Feb '16

Young Lee will present his FPO, "Targeted analyses of very large genome-wide data collections" on Tuesday, 3/1/2016 at 10:30am in CS 402 The members of his committee are Olga Troyanskaya (adviser), readers: Mona Singh and Barbara Engelhardt; nonreaders: John Storey (Lewis-Sigler Institute for Integrative Genomics) and Tom Funkhouser. A copy of his thesis, is available in Room 310. Everyone is invited to attend his talk. The talk title and abstract follow below. "Targeted analyses of very large genome-wide data collections" Abstract: Genome-scale experiments provide an overwhelming amount of molecular information for biologist. New computational methods are needed for specific analysis and interpretation of such high-dimensional data. Here we take advantage of the massive public repositories to quantify the tissue-specific signals in gene expression profiles, characterize distinctive molecular features of human diseases, deconvolve the latent cell-type-specific factors in mixed clinical samples, and automatically integrate heterogeneous data sources in the context of a specific genome-wide dataset. First, we describe URSA (Unveiling RNA Sample Annotation) that incorporates the known tissue/cell-type relationships to better estimate the specific signal in any given gene expression profile. Our ontology-aware method combines independent discriminative classifiers in a Bayesian framework, outperforming other machine learning methods. We provide a molecular interpretation for the tissue and cell-type models learned by URSA, enabling a data-driven view of molecular processes specific to particular tissues and cell types. Then, we extend this work for human diseases. We use thousands of clinical disease-specific expression profiles in public repositories to quantify distinctive functional and anatomical characteristics of human diseases. Through our data-driven analysis, we explore the complexity of the human disease landscape and propose exploratory hypothesis for drug repurposing even for rare disease with no prior genetic knowledge. Lastly, we describe YETI (Your Evidence Tailored Integration) for targeted integration of heterogeneous genome-wide data sources. Biomedical researchers generate genome-wide datasets for data-driven exploration of specific questions but such analyses are disconnect from big public data collections. YETI is the first automatic integration method that effectively constructs functional networks specific to a genome-scale dataset. We show that the resulting integration reflect the biological context of the user-provided dataset while providing accurate prediction for functional interactions.

1 0