Skip to content. Skip to navigation
McGill Home SOCS Home
Personal tools
You are here: Home Announcements and Events Colloquium Archive

Colloquium Home
Fall 2014 Schedule
Winter 2015 Schedule
How to Suggest a Speaker
Archives

 

Date
( Winter 2009 )
Speaker and Abstract
2009/01/09 Speaker: Marcello Pelillo
Affiliation: Ca' Foscari University, Venice, Italy
Title: From Cliques to Equilibria: The Dominant-Set Framework for Pairwise Data Clustering
Abstract: In this talk I'll provide an overview of the work we have been doing in my group on pairwise data clustering. We have developed a graph-theoretic framework for pairwise clustering which is motivated by the analogies between the intuitive concept of a cluster and that of a "dominant set" of vertices, a novel notion that generalizes that of a maximal clique to edge-weighted graphs. We established a correspondence between dominant sets and the extrema of a quadratic form over the standard simplex, thereby allowing us to use straightforward continuous optimization techniques such as replicator dynamics from evolutionary game theory. We have also addressed the problem of grouping out-of-sample examples after the clustering process has taken place: as it turns out, the very notion of a dominant set offers a simple and efficient way of doing this. Finally, we have provided a generalization of the dominant-set notion to the case of arbitrary (i.e., negative and/or asymmetric) affinities which is based on the notion of a Nash equilibrium, a fundamental concept fron noncooperative game theory. The approach has been tested extensively on on image segmentation and perceptual grouping problems. [joint work with M. Pavan, A. Torsello, S. Rota Bulo'] Biography of Speaker:

Marcello Pelillo is an associate professor of computer science at Ca' Foscari University, Venice, where he also serves as the chair of the board of studies of the Computer Science School. He held visiting research positions at Yale University, the University College London, McGill University, the University of Vienna, and York University (England). He has published more than 100 technical papers in refereed journals, handbooks, and conference proceedings in the areas of computer vision, pattern recognition and neural computation. He was actively involved in the organization of several scientific meetings and, in 1997, he co-established a new series of international workshops devoted to energy minimization methods in computer vision and pattern recognition (EMMCVPR), which has now reached the sixth edition www.emmcvpr org. He is regularly on the program committees of the major international meetings in his fields and serves on the editorial board for the journals IEEE Transactions on Pattern Analysis and Machine Intelligence and Pattern Recognition. Prof. Pelillo is a Senior Member of the IEEE and a Fellow of the IAPR.


2009/01/16 Speaker: Ashish Sabharwal
Affiliation: Cornell University
Area: AI
Title: A Quest for Large-Scale Reasoning and Sampling Methods
Abstract: A key goal of Artificial Intelligence is to develop scalable and robust automated reasoning technology that will allow computers to act intelligently in increasingly complex real-world settings and in competitive and uncertain environments. One of the most successful stories in this quest has been that of combinatorial reasoning, in particular for propositional satisfiability (SAT) problems and constraint reasoning. Starting with only a few hundred variable problems in the 1990's, we can now solve interesting problems with over a million variables. This talk will discuss the first formal study of learning methods used in today's highly engineered inference engines in terms of resolution complexity, and an effective practical approach to exploit a problem's natural symmetric structure by going beyond the limits inherent in resolution-based inference. We will then touch upon new methods for more complex reasoning problems, namely counting and sampling solutions as well as clusters of solutions. These new methods have pushed the limits of scalability by orders of magnitude, and promise to open up a range of new applications in AI and computer science in general, particularly those involving the integration of logical and probabilistic inference. Biography of Speaker:

Ashish Sabharwal is a Research Associate in Computer Science at Cornell University, and works at the Intelligent Information Systems Institute and the Institute for Computational Sustainability. He received his M.S. and Ph.D. from the University of Washington, Seattle in 2005, and was a Postdoctoral Associate at Cornell till 2008. His research interests include combinatorial methods, probabilistic inference, multi-agent reasoning, connections to statistical physics, game theory, algorithm design, and complexity. He has (co-)authored over 30 publications and surveys, including two Best Paper Awards, one runner-up prize, and four best paper nominations.


2009/03/13 Speaker: Bohdan Kaluzny
Affiliation: Centre for Operational Research & Analysis, Defence Research & Development Canada:
Area: Theory
Title: Combinatorial Optimization in Airlift Modeling
Abstract: The Canadian Forces make extensive use of airlift assets to deploy and sustain military and humanitarian operations overseas. Various-sized cargo aircraft are employed for missions to deliver equipment, supplies, and passengers. Minimizing the number of trips (sorties) required is synonymous with maximizing the payload delivered per flight while respecting constraints on maintaining a safe load balance. This talk presents Genetic Annealing for Loading of Aircraft: a Heuristic Aiding Deployment (GALAHAD). This loading module uses a genetic algorithm based on a bin packing heuristic and a convex hull fitness function to generate realistic estimates of the number of airlift sorties required to transport a given manifest of items. In addition to standard genetic operators, a novel extinction phase is employed. Load balancing of individual sorties are modeled with Mixed Integer Linear Programs and an extinction is invoked sparingly (due to computational time) during the genetic algorithm to remove solutions which contain infeasible allocations of items to an airlift asset. The program returns the Pareto set of optimal solutions (number of sorties of different aircraft types) that can then be further evaluated by decision makers. Biography of Speaker:

Bohdan Kaluzny is a Defence Scientist at the Centre for Operational Research & Analysis, Defence Research & Development Canada. He received his B.Sc. (1999), M.Sc. (2001) and Ph.D. (2005) from the School of Computer Science, McGill University. His research interests include combinatorial optimization, algorithm design, multi-criteria decision analysis and reliability analysis. He is currently providing scientific support to the Department of National Defence Major Project Delivery Division which overseas multi-million/billion dollar Canadian Forces acquisition projects such as arctic patrol ships, search and rescue helicopters, Canadian Navy surface combatants, logistic support vehicles, etc.


2009/03/26 Speaker: Philippe Cudre-Mauroux
Affiliation: MIT
Area: bioinfo
Title:
Abstract: Biography of Speaker:

Philippe Cudre-Mauroux


2009/03/27 Speaker: Sylvain Bouix
Affiliation: Harvard University
Title: An Introduction to Neuroimaging Research in Schizophrenia
Abstract: MR imaging has revolutionized the way neuroscientist study the brain, as it has allowed us to investigate the morphology and function of the brain /in vivo/, without any harming effects to human subjects. This talk will give an overview of current state of the art imaging research in Schizophrenia. We will discuss how structural, diffusion MRI analysis can be used to get a better understanding of the brain abnormalities involved in schizophreni Biography of Speaker:

Sylvain Bouix is an Instructor in Psychiatry at Harvard Medical School and Associate Director of the Psychiatry Neuroimaging Laboratory, Department of Psychiatry, Brigham and Women's Hospital where he acts as a mediator between computer scientists and neuroscientists to help improve computer tools for neuroimaging studies. He received his B.Eng. from Institut Polytechnique de Sevenans , France, his M.Sc. from the University of Kansas , and his Ph.D. in Computer Science from McGill University . His research interests include shape analysis of anatomical structures and evaluation of medical imaging techniques.


2009/04/03 Speaker: Nicholas Roy
Affiliation: MIT
Title: Planning and Learning in Information Space
Abstract: Decision making with imperfect knowledge is an essential capability for unmanned vehicles operating in populated, dynamic domains. For example, a UAV flying autonomously indoors will not be able to rely on GPS for position estimation, but instead use on-board sensors to track its position and map the obstacles in its environment. The planned trajectories for such a vehicle must therefore incorporate sensor limitations to avoid collisions and to ensure accurate state estimation for stable flight -- that is, the planner must be be able to predict and avoid uncertainty in the state, in the dynamics and in the model of the world. Incorporating uncertainty requires planning in information space, which leads to substantial computational cost but allows our unmanned vehicles to plan deliberate sensing actions that can not only improve the state estimate, but even improve the vehicle's model of the world and how people interact with the vehicle. I will discuss recent results from my group in planning in information space; our algorithms allow robots to generate plans that are robust to state and model uncertainty, while planning to learn more about the world. I will describe the navigation system for a quadrotor helicopter flying autonomously without GPS using laser range-finding, and will show how these results extend to autonomous mapping and human-robot interaction. Biography of Speaker:

Nicholas Roy is the Boeing Assistant Professor in the Department of Aeronautics & Astronautics at the Massachusetts Institute of Technology and a member of the Computer Science and Artificial Intelligence Laboratory (CSAIL) at MIT. He received his B.Sc. and M.Sc. from McGill University, and his Ph. D. in Robotics from Carnegie Mellon University in 2003. His research interests include autonomous systems, mobile robotics, human-computer interaction, decision-making under uncertainty and machine learning.


2009/04/10 Speaker: Lui Sha
Affiliation: University of Illinois at Urbana Champaign
Title: Overview of the US Cyber-Physical Systems Initiative
Abstract: The Internet has made the world "flat" by transcending space. We can now interact with people and get useful information around the globe in a fraction of a second. The Internet has transformed how we conduct research, studies, business, services, and entertainment. However, there is still a serious gap between the cyber world, where information is exchanged and transformed, and the physical world in which we live. The emerging cyber-physical systems shall enable a modern grand vision for societal-level services that transcend space and time at scales never possible before. CPS is now the top priority of US IT research and I will give an overview of the CPS challenges presented in NSF CPS research planning workshops. Biography of Speaker:

Lui Sha graduated with BSEE from McGill in 1978 and graduated with Ph.D. from Carnegie Mellon University in 1985. Currently, he is Donald B. Gillies Chair professor of Computer Science at University of Illinois at Urbana Champaign. He is a fellow of the ACM and IEEE. His work transformed the IEEE standards on real time computing and was cited as a major accomplishment of computer science in National Academy of Science's report: A Broader Agenda for Computer Science and Engineering. His work was also cited as a key enabling technology for many high technology projects, including GPS in orbit upgrade, the Mars Pathfinder and the International Space Station. He served on US National Academy of Science's study committee on certifiably dependable software, and is a member of the organizing committee for CPS Research Initiative


2009/04/24 Speaker: Oussama Khatib
Affiliation: Stanford University
Area: Robotics
Title: Human-Friendly Robotics
Abstract: Robotics is rapidly expanding into the human environment and vigorously engaged in its new emerging challenges. Interacting, exploring, and working with humans, the new generation of robots will increasingly touch people and their lives. The successful introduction of robots in human environments will rely on the development of competent and practical systems that are dependable, safe, and easy to use. This presentation focuses on our ongoing effort to develop human-friendly robotic systems that combine the essential characteristics of safety, human-compatibility, and performance. In human-friendly robot design, our effort has focused on new design concepts for the development of intrinsically safe robotic systems that possess the requisite capabilities and performance to interact and work with humans. The result is a novel hybrid actuation approach that combines the use of small motors at the joints with pneumatic, muscle-like actuators remotely connected by cables. With this hybrid actuation, the impedance of the resulting robot is decreased by an order of magnitude, making it substantially safer without sacrificing performance. To develop improved skills for human robot interaction, we pursued an extensive study of human motion to unveil its underlying characteristics, and to formulate general strategies for interactive whole-body robot control. Our exploration has employed models of human musculoskeletal dynamics and used experimental studies of human subjects with motion capture techniques. This investigation has revealed the dominant role physiological characteristics play in shaping human motion. Using these characteristics we develop generic motion behaviors that efficiently and effectively encode some basic human motion behaviors. The implementation of these behaviors on robots with complex human-like structures relied on a unified whole-body task-oriented control structure that addresses dynamics in the context of multiple tasks, multi-point contacts, and multiple constraints. The performance and effectiveness of this approach are demonstrated through extensive robot dynamic simulations and implementations on physical robots for experimental validation. Biography of Speaker:

Dr. Khatib is Professor of Computer Science at Stanford University. He received his Ph.D. in 1980 from Sup'Aero, Toulouse, France. His current research is in human-centered robotics, haptic interactions, and human-friendly robot design. Professor Khatib was the Program Chair of ICRA2000 (San Francisco) and Editor of ``The Robotics Review'' (MIT Press). He served as the Director of the Stanford Computer Forum, an industry affiliate program. He is the President of the International Foundation of Robotics Research, IFRR, Editor of STAR, Springer Tracts in Advanced Robotics, and Editor of Springer Handbook of Robotics. Professor Khatib is an IEEE Fellow who served as a Distinguished Lecturer of IEEE, and is a recipient of the JARA Award.