Conferences, Lectures, & Seminars
Events for February
-
Predicting Human Body Shape Under Clothing
Mon, Feb 02, 2009 @ 03:30 PM
Thomas Lord Department of Computer Science
Conferences, Lectures, & Seminars
Speaker: Prof. Michael J. Black, Brown University
Host: Prof. Gerard MedioniAbstract:
We propose a method to estimate the detailed 3D shape of a person from images of that person wearing clothing. The approach exploits a model of human body shapes that is learned from a database of over 2000 range scans. We show that the parameters of this shape model can be recovered independently of body pose. We further propose a generalization of the visual hull to account for the fact that observed silhouettes of clothed people do not provide a tight bound on the true 3D shape. With clothed subjects, different poses provide different constraints on the possible underlying 3D body shape. We consequently combine constraints across pose to more accurately estimate 3D body shape in the presence of occluding clothing. Finally we use the recovered 3D shape to estimate the gender of subjects and then employ gender-specific body models to refine our shape estimates. Results on a novel database of thousands of images of clothed and ``naked'' subjects, as well as sequences from the HumanEva dataset, suggest the method may be accurate enough for biometric shape analysis in video.This is joint work with Alexandru Balan. Project page: http://www.cs.brown.edu/~alb/scapeClothing/Related ECCV paper: http://www.cs.brown.edu/~black/Papers/balanECCV08.pdfBiography:
Michael Black received his B.Sc. from the University of British Columbia (1985), his M.S. from Stanford (1989), and his Ph.D. in computer science from Yale University in 1992. He has been a visiting researcher at the NASA Ames Research Center and an Assistant Professor in the Dept. of Computer Science at the University of Toronto. In 1993 Prof. Black joined the Xerox Palo Alto Research Center where he managed the Image Understanding area and later founded the Digital Video Analysis group. In 2000, Prof. Black joined the faculty of Brown University where he is a Professor of Computer Science. At CVPR'91 he received the IEEE Computer Society Outstanding Paper Award for his work with P. Anandan on robust optical flow estimation. His work also received Honorable Mention for the Marr Prize in 1999 (with David Fleet) and 2005 (with Stefan Roth). Prof. Black's research interests in machine vision include optical flow estimation, human motion analysis and probabilistic models of the visual world. In computational neuroscience his work focuses on probabilistic models of the neural code, the neural control of movement and the development of neural interface systems that directly connect brains and machines to restore lost function to people with central motor system injury.Location: Seaver Science Library (SSL) - 150
Audiences: Everyone Is Invited
Contact: CS Colloquia
-
The Optimization and Economic Aspects of Internet Advertising
Thu, Feb 05, 2009 @ 04:00 PM
Thomas Lord Department of Computer Science
Conferences, Lectures, & Seminars
Speaker: Hamid Nazerzadeh, Stanford University
Host: Prof. David KempeAbstract:
In the last few years, Internet advertising has grown to a multi-billion dollar market, and has become the dominant business model for the companies that provide content and services online. Hence, there has been a surge of interest in the problems emerging in the design and implementation of these markets. In this talk, I will discuss some of these problems with a focus on sponsored search, the auctions run by search engines such as Google, Yahoo!, and MSN to sell advertisement space alongside the search results. In this context, I will talk about online allocation of advertisement space and designing cost-per-action mechanisms.Biography:
Hamid Nazerzadeh is a Ph.D. student in operations research at Stanford university, working under the supervision of Amin Saberi and Ashish Goel. His research interests lie at the intersection of operations research, computer science, and economics. His dissertation work is mainly focused on the optimization and economic aspects of Internet advertising. He received a Yahoo! Ph.D. student fellowship in 2007.Location: Seaver Science Library (SSL) - 150
Audiences: Everyone Is Invited
Contact: CS Colloquia
-
Explaining Market Price Discovery as an Algorithmic Process
Thu, Feb 12, 2009 @ 04:00 PM
Thomas Lord Department of Computer Science
Conferences, Lectures, & Seminars
Speaker: Prof. Richard Cole, New York University
Host: Prof. David KempeAbstract:
Self-organizing behavior can often be characterized in terms of a distributed process. It is natural to ask when and why it arises.One instance of such a distributed process is pricing in markets. A basic tenet of well-functioning markets is that they discover (converge toward) prices that simultaneously balance supplies and demands of all goods; these are called equilibrium prices. Further, the markets are self-stabilizing, meaning that they converge toward new equilibria as conditions change. This talk will seek to explain why this might happen by taking an algorithmic approach.More specifically, we introduce the setting of Ongoing Markets (by contrast with the classic Fisher and Exchange markets). An Ongoing Market allows trade at non-equilibrium prices, and, as its name suggests, continues over time. The main task remaining is to specify and analyze a price update rule. We consider a (tatonnement-style) rule with the following characteristics:1. The procedure is distributed: (i) the price updates for each good are independent, and (ii) the update for each good uses only limited "local" information. 2. It is asynchronous: price updates do not have to be simultaneous. 3. It is simple.And for appropriate markets the rule enables:4. Fast convergence. 5. Robustness in the face of (somewhat) inaccurate data.This talk is intended for a general (Computer Science) audience; it is based on joint works with Lisa Fleischer and Ashish Rastogi.Biography:
Richard Cole is a professor of Computer Science in the Courant Institute at NYU, where he has been on the faculty since receiving his Ph.D. in 1982. His Ph.D. was from Cornell, supervised by John Hopcroft. He served as department chair from 1994-2000. He was a fellow of the Guggenheim Foundation in 1988-89, and was named an ACM Fellow in 1998. He is the author or coauthor of over 100 papers. Highlights include the Parallel Merge Sort algorithm, the proof of the Dynamic Finger Conjecture for Splay Trees, and a tight analysis of the Boyer-Moore string matching algorithm.Location: Seaver Science Library (SSL) - 150
Audiences: Everyone Is Invited
Contact: CS Colloquia
-
New Temporal-Difference Methods Based on Gradient Descent
Wed, Feb 18, 2009 @ 04:00 PM
Thomas Lord Department of Computer Science
Conferences, Lectures, & Seminars
Speaker: Prof. Richard Sutton, University of Alberta
Host: Prof. Stefan Schaal Abstract:
Temporal-difference methods based on gradient descent and parameterized function approximators form a core part of the modern field of reinforcement learning and are essential to many of its large-scale applications. However, the most popular methods, including TD(lambda), Q-learning, and Sarsa, are not true gradient-descent methods and, as a result, the conditions under which they converge are narrower and less robust than can usually be guaranteed for gradient-descent methods. In this paper we introduce a new family of temporal-difference algorithms whose expected updates are in the direction of the gradient of a natural performance measure that we call the "mean squared projected Bellman error". Because these are true gradient-descent methods, we are able to apply standard techniques to prove them convergent and stable under general conditions including, for the first time, off-policy training. The new methods are of the same order of complexity as TD(lambda) and, when TD(lambda) converges, they converge at a similar rate to the same fixpoints. The new methods are similar to GTD(0) (Sutton, Szepesvari & Maei, 2009), but based on a different objective function and much more efficient, as we demonstrate in a series of computational experiments. (this is joint work with Hamid Maei, Doina Precup, Csaba Szepesvari, Shalabh Bhatnagar, David Silver, and Eric Wiewiora) Biography:
Richard S. Sutton is a professor and iCORE chair in the department of computing science at the University of Alberta. He is a fellow of the Association for the Advancement of Artificial Intelligence and co-author of the textbook Reinforcement Learning: An Introduction from MIT Press. Before joining the University of Alberta in 2003, he worked in industry at AT&T and GTE Labs, and in academia at the University of Massachusetts. He received a PhD in computer science from the University of Massachusetts in 1984 and a BA in psychology from Stanford University in 1978. Rich's research interests center on the learning problems facing a decision-maker interacting with its environment, which he sees as central to artificial intelligence. He is also interested in animal learning psychology, in connectionist networks, and generally in systems that continually improve their representations and models of the world.Location: Ronald Tutor Hall of Engineering (RTH) - 406
Audiences: Everyone Is Invited
Contact: CS Colloquia
-
Learning Similarities and Dimensionality Reduction
Thu, Feb 19, 2009 @ 04:00 PM
Thomas Lord Department of Computer Science
Conferences, Lectures, & Seminars
Speaker: Dr. Kilian Weinberger, Yahoo! Research
Host: Prof. Fei ShaAbstract:
One of the most fundamental challenges of machine learning and artificial intelligence is the learning of suitable representations of data. Many machine learning algorithms assume that the data is presented in low dimensional vectorial form, where Euclidean distances reflect dissimilarities. Often this raw data format is far from optimal. Ideally one should be able to learn a "hand-tailored" representation of each particular data set for any given task. In this talk, I present three algorithms for learning compact representations that give rise to semantically meaningful similarity metrics. Each of the algorithms involves, at its core, a convex optimization problem that learns the new representation under meaningful constraints. This framework provides perfect reproducibility and theoretical guarantees. The three methods are most suitable for different data settings: Maximum Variance Unfolding reduces the dimensionality of data sets with underlying manifold structure. Taxonomy Embedding is a powerful tool for hierarchical document categorization. Large Margin Nearest Neighbor learns a robust metric for k-nearest neighbor classification. I present state-of-the-art classification results on several real world applications, including handwritten digit recognition on the MNIST corpus and document categorization on the OHSUMED medical journal data base. Biography:
Kilian Weinberger is a Research Scientist at Yahoo Research in Santa Clara, California. He works on next-generation spam filtering algorithms, multimedia search and machine learning with convex optimization. In 2007 he received a Ph.D. in Computer Science at the University of Pennsylvania under the supervision of Prof. Lawrence Saul. His work on supervised and unsupervised metric learning won several outstanding paper awards at CVPR, AISTATS and ICML. Prior to his doctoral studies he earned a first class honor BA in Mathematics and Computer Science from the University of Oxford.Location: Seaver Science Library (SSL) - 150
Audiences: Everyone Is Invited
Contact: CS Colloquia
-
Computational Study Of Nonverbal Social Communication
Thu, Feb 26, 2009 @ 04:00 PM
Thomas Lord Department of Computer Science
Conferences, Lectures, & Seminars
Speaker: Dr. Louis-Philippe Morency, ICT, USC
Host: Prof. Stefan SchaalAbstract:
The goal of this emerging research field is to recognize, model and predict human nonverbal behavior in the context of interaction with virtual humans, robots and other human participants. At the core of this research field is the need for new computational models of human interaction emphasizing the multi-modal, multi-participant and multi-behavior aspects of human behavior. This multi-disciplinary research topic overlaps the fields of multi-modal interaction, social psychology, computer vision, machine learning and artificial intelligence, and has many applications in areas as diverse as medicine, robotics and education. During my talk, I will focus on three novel approaches to achieve efficient and robust nonverbal behavior modeling and recognition: (1) a new visual tracking framework (GAVAM) with automatic initialization and bounded drift which acquires online the view-based appearance of the object, (2) the use of latent-state models in discriminative sequence classification (Latent-Dynamic CRF) to capture the influence of unobservable factors on nonverbal behavior and (3) the integration of contextual information (specifically dialogue context) to improve nonverbal prediction and recognition.Biography:
Dr. Louis-Philippe Morency is currently research scientist at USC Institute for Creative Technologies where he leads the Nonverbal Behaviors Understanding project (ICT-NVREC). He received his Ph.D. from MIT Computer Science and Artificial Intelligence Laboratory in 2006. His main research interest is computational study of nonverbal social communication, a multi-disciplinary research topic that overlays the fields of multi-modal interaction, computer vision, machine learning, social psychology and artificial intelligence. He developed "Watson", a real-time library for nonverbal behavior recognition and which became the de-facto standard for adding perception to embodied agent interfaces. He received many awards for his work on nonverbal behavior computation including three best-paper awards in 2008 (at various IEEE and ACM conferences). He was recently selected by IEEE Intelligent Systems as one of the "Ten to Watch" for the future of AI research.Location: Seaver Science Library (SSL) - 150
Audiences: Everyone Is Invited
Contact: CS Colloquia
