April 30, 2012
Modeling Rich Structured Data via Kernel Distribution Embeddings
By
Dr. Le Song (Georgia Tech, USA)
In Real world applications often produce a large volume of highly uncertain and complex data. Many of them have rich microscopic structures where each variable can take values on manifolds (e.g., camera rotations), combinatorial objects (e.g., texts, graphs of drug compounds) or high dimensional continuous domains (e.g., images and videos). Furthermore, these problems may possess additional macroscopic structures where the large collections of observed and hidden variables are connected by networks of conditional independence relations (e.g., in predicting depth from still images, and forecasting in time-series).
Most previous learning algorithms for problems with such rich structures rely heavily on linear relations and parametric models where data are typically assumed to be multivariate Gaussian or discrete with a relatively small number of values. Conclusions inferred under these restricted assumptions can be misleading, if the underlying data generating processes contain nonlinear, non-discrete, or non-Gaussian components.
How can we find a suitable representation for nonlinear and non-Gaussian relationships in a data-driven fashion? How can we exploit conditional independence structures between variables in rich structured setting? How can we design efficient algorithms to solve challenging nonparametric problems involving large amount of data?
In this talk, I will introduce a nonparametric representation for distributions called kernel embeddings to address these questions. The key idea of the method is to map distributions to their expected features (potentially infinite dimensional), and given evidence, update these new representations solely in the feature space. Compared to existing nonparametric representations which are largely restricted to vectorial data and usually lead to intractable algorithms, very often kernel distribution embeddings lead to simpler, faster and more accurate algorithms in a diverse range of problems such as predicting depth from still images and forecasting in time-series.
Biography: Prof Le Song studied computer science at the South China University of Technology, Guangzhou, China in 1998. After he obtained his Bachelor's degree in 2002, he traveled to Sydney, Australia. In 2004 he received my Master's degree, and in 2008 his Doctoral degree in computer science at the University of Sydney, Australia. He was also a PhD. student at the Statistical Machine Learning Program at NICTA. Since Summer 2008, he was a Lane postdoc fellow at Carnegie Mellon University, working on machine learning and computational biology projects with Eric Xing, Carlos Guestrin, Geoff Gordon and Jeff Schneider. Right before he joining Georgia Tech, he spent some time as a research scientist in Fernando Pereira's group at Google Research.
| Date & Time: Monday, April 30th, 2012 from 03:00 pm to 04:00 pm Location: Room # 2418, level 2, building 1 (Al-Khwarizmi) Refreshments: Available in 3119 @ 02:45 pm |
April 28, 2012
Does Your Computer Know How To Add
By
Dr. Hossam A.H. Fahmy (Cairo University, Egypt)
In this talk a brief introduction to computer arithmetic is provided and the topic of errors in floating point calculations is exposed. The hardware and SW support for correct functionality are discussed and the current research work in decimal floating point is described.
Biography: Hossam A. H. Fahmy earned his PhD from the Electrical Engineering Department of Stanford University, USA in 2003. He received his BSc and MSc degrees from the Electronics and Communications Engineering Department of Cairo University, Egypt in 1995 and 1997 respectively. Currently, he is an Associate Professor with the same department where he leads research on computer arithmetic, computer architecture, and computer fonts. He has taught previously at the German University in Cairo, the American University in Cairo, Nanyang Technological University in Singapore, and King Abdullah University for Science and Technology in Saudi Arabia.
Dr. Fahmy's previous research include collaborations with IMEC in Belgium, IBM T.J. Watson research center in USA, and AMD in California, USA. He served as a member of the working group developing the IEEE standard 754-2008 on Floating Point Arithmetic from 2002 till its publication in 2008. He is currently serving as a member of the working group formulating the future IEEE standard P1788 on Interval Arithmetic. Dr. Fahmy is co-founder and Chief Technical Officer of SilMinds the world leader in providing hardware decimal floating point units.
Date & Time: Saturday, April 28th, 2012 from 02:00 pm to 03:00 pm
Location: Room # 3119, level 3, building 1 (Al-Khwarizmi)
Refreshments: Available in 3119 @ 01:45 pm
April 28, 2012
The Wonders of Electron Density: from Fractional Charges and Fractional Spins to the Non-Covalent Interactions of Bio-Molecular Complexes
By
Dr. Weitao Yang (Duke University)
Interactions between electrons determine the structure and properties of matter from molecules to solids. To describe interacting electrons, the extremely simple three-dimensional electron density can be used as the basic variable within density functional theory, negating the need in many cases for the massively complex many-dimensional wave function. This lecture will review the concept of electron density and many of its remarkable features, including latest developments. Electron density determines the energetics of chemical and physical changes, such as in electron transport, solar energy harvesting and conversion, drug design in medicine, catalytic processes in enzymes, and many other challenges in science and technology. Electron density can be used to visualize not only covalent chemical bonds, but also non-covalent interactions such as hydrogen bonds, van der Waals attraction and steric repulsion. Electron density also naturally leads to the concepts of fractional charges and fractional spins, which are key to the understanding and analysis of approximations in density functional theory.
References
1. J. Cohen, P. Mori-Sanchez, and W. T. Yang. Insights into current limitations of density functional theory. Science, 321:792, 2008.
2. P. Mori-Sanchez, A. J. Cohen, and W. T. Yang. Localization and delocalization errors in density functional theory and implications for band-gap prediction. Physical Review Letters, 100(14):146401, 2008.
3. P. Mori-Sanchez, A. J. Cohen, and W. T. Yang. Discontinuous Nature of the Exchange-Correlation Functional in Strongly Correlated Systems, Physical Review Letters, 102:066403, 2009.
4. E. R. Johnson, S. Keinan, P. Mori-Sanchez, J. Contreras-Garcia, A. J. Cohen, and W. T. Yang. Revealing noncovalent interactions. Journal of the American Chemical Society, 132(18):6498-6506, 2010.
5. X. Zheng, A. J. Cohen, P. Mori-Sanchez, X. Q. Hu, and W. T. Yang. Improving band gap prediction in density functional theory from molecules to solids. Physical Review Letters, 107(2):026403, 2011.
6. J. Cohen, P. Mori-Sanchez, and W. T. Yang. Challenges for Density Functional Theory. Chem. Rev. 112:289, 2012
Biography: Prof. Yang received his PhD in chemistry from University of North Carolina, Chapel Hill in 1986 following a BSc in chemistry at Peking University. After worked at UC Berkeley as a research associate, he joined Duke University in 1989 and became the Philip Handler Professor of Chemistry in 2003. He is an elected fellow of both AAAS (American Association for the Advancement of Science) and APS (American Physical Society). His research interests are physical chemistry, theoretical chemistry, electronic structure theory, density functional theory, biophysical chemistry, nano science and condensed matters.
http://fds.duke.edu/db/aas/Chemistry/faculty/weitao.yang
Date & Time: Saturday, April 28, 2012 from 10:30 to 11:30 AM
Location: Room # 3119, level 3, building 1, Al Khwarizmi
Refreshments: Available in 3119 @ 10:15 PM
April 25, 2012
Investigating the Fundamentals of Combustion using Implicit Simulation Method with Realistic Chemistry
By
Dr. Nadeem A. Malik (KFUPM)
Although implicit methods have a history in the combustion community, they have not been favored due to large time and memory requirements. However, the tide may be turning as the stability of implicit methods is essential for the inclusion of realistic chemistry which is required for more accurate predictive methods. A recent method TARDIS (Transient Advection Reaction Diffusion Implicit Simulations), [1-3], successfully couples the compressible flow to the comprehensive chemistry and multi-component transport properties; TARDIS has been demonstrated in application to two fundamental combustion problems of great interest, namely in flame-pressure interactions in transient premixed flames, and in the study of stretched laminar flame velocities.
The author will review the TARDIS implicit method and its recent applications in computational combustion, and the formidable challenges that still need to be met in order to reduce the costs of running large implicit simulations in 2D and 3D.
[1] Malik, N.A. and Lindstedt, R.P. The response of transient inhomogeneous flames to pressure fluctuations and stretch: planar and outwardly propagating hydrogen/air flames. Combust. Sci. Tech. 82(9), 2010.
[2] Malik, N. A. “Non-linear powers laws in stretched flame velocities in finite thickness flames: a numerical study using realistic chemistry”. Combust. Sci. Tech. Accepted, to appear, 2012.
[3] Malik, N.A. and Lindstedt, R.P. The response of transient inhomogeneous flames to pressure fluctuations and stretch: planar and outwardly propagating hydrogen/air flames. Combust. Sci. Tech. Accepted, To appear, 2012.
Biography: Nadeem Malik earned his PhD in Mathematics from Cambridge University in 1991 following a BSC in Mathematics at Imperial College. He is an Associate Professor in the Department of Mathematics and Statistics at King Fahd University of Petroleum and Minerals following earlier post-doctoral, research, and teaching appointments at ETH-Zurich, Ecole Centrale de Lyon, Imperial College, and Queen Mary University. His current interests are computational combustion, turbulent diffusion, transport through porous media, and particle tracking. Web: http://www.kfupm.edu.sa/math/namalik and http://www.linkedin.com/in/nmalik11
Date & Time: Wednesday, April 25, 2012 from 03:30 to 04:30 PM
Location: Room # 3119, level 3, building 1, Al Khwarizmi
Refreshments: Available in 3119 @ 03:15 PM
April 15, 2012
Analysis of an Alternating Regularisation Technique for PET Reconstruction
By
Dr. Carola-Bibiane Schonlieb (Cambridge University)
We discuss a recently proposed reconstruction technique for positron emission tomography (PET) and present its well-posedness analysis. In contrast to the majority of existing algorithms which apply denoising to the reconstructed image, our new approach applies a regularisation both in the measurement and the image space. For this task we use an alternating total variation algorithm. This is joint work with E. Papoutsellis and is based on the model proposed in P. E. Barbano, T. Fokas and C.-B. Schönlieb.
Biography: Carola-Bibiane Schönlieb is a lecturer in Applied and Computational Mathematics in the Department of Applied Mathematics and Theoretical Physics at Cambridge University, and a Fellow of Jesus College, Cambridge. Her research interests range from nonlinear partial differential equations to computational- and convex analysis, with applications in digital image- and signal processing. She studies fourth-order equations and nonsmooth optimization problems, like the total variation functional, for image reconstruction, especially for what is called image inpainting. Moreover, she works on computational methods for large-scale problems appearing in 3- and 4-D imaging. Within this context she is interested in both the theoretical and numerical analysis of the problems considered as well as their practical implementation and their use for real-world applications like arts restoration and medical imaging. Dr. Schönlieb holds an M.A. in mathematics from the University of Salzburg and a Ph.D. in mathematics from the University of Cambridge.
Date & Time: Sunday, April 15, 2012 from 03:00 to 04:00 PM
Location: Room # 2418, level 2, area 2, building 1, Al Khwarizmi
Refreshments: Available in 2418 @ 02:45 PM
April 10, 2012
Development and Application of a Massively Parallel Multi-Scale, Multi-Phase, Multi-Component Groundwater Model
By
Dr. Jitendra Kumar (Oak Ridge National Laboratory)
Numerical modeling is a critical tool for evaluation and design of remediation strategies at legacy waste site, geologic carbon sequestration and to understand the role of subsurface hydrology for regional and global climate system. Physical and chemical complexity of the site applications pose challenges to the state of the art subsurface models. Representation of highly heterogeneous stratified, fractured and weathered aquifers, multi-component geochemical and biological reactions and transport, multi-phase flows are some of the issues which requires high resolution (spatial and temporal), multi-scale models with detailed mechanistic representation of subsurface flow and transport processes. These complex problems often involve billions of degrees of freedoms and require use of state of the art high performance computing resources and techniques. Application to real world sites also requires computationally intensive model-data integration and inverse modeling to assimilate field and laboratory based observations in the simulation models.
I will present some of my work and experiences on development of new process modules in massively parallel reactive flow and transport model PFLOTRAN, its application to Oak Ridge Integrated Field Research Challenge site and model-data integration using a parallel inverse modeling framework. I will also present some recent work on parallel data mining algorithms and data analytics for very large Earth Science datasets.
Biography: Jitendra Kumar is a postdoctoral research associate in the Computer Earth Sciences Group at Oak Ridge National Laboratory (ORNL). His primary research interests are in modeling of surface/subsurface hydrologic flow and transport processes, their interactions and feedbacks with the terrestrial biosphere, data mining algorithms and data analytics for large data sets, environmental systems analysis, and high performance computing. Jitendra received his PhD in Civil Engineering from North Carolina State University. For his dissertation he developed and implemented a parallel evolutionary algorithm based simulation-optimization framework for contaminant warning systems in water distribution networks. He received his MS in Hydraulics and Water Resources Engineering from Indian Institute of Technology Kanpur, India. He conducted his MS Thesis research at Technische Universitat Darmstadt, Germany under a Indo-German student exchange scholarship. He also holds a B.S degree in Civil Engineering from Muzaffarpur Institute of Technology, India.
Date & Time: Tuesday, April 10, 2012 from 10:00 am to 11:00 am
Location: Room # 3119, level 3, building 1, Al Khwarizmi
Refreshments: Available in 3119 @ 09:45 am
April 09, 2012
The Navier-Stokes, Euler and Other Related Equations
By
Dr. Titi (UC Irvine, USA)
In this talk I will survey the status of, and the most recent advances concerning, the questions of global regularity of solutions to the three-dimensional Navier-Stokes and Euler equations of incompressible fluids. Furthermore, I will also present recent global regularity results concerning certain three-dimensional geophysical fows, including the three-dimensional viscous “primitive equations" of oceanic and atmospheric dynamics.
Biography: Professor Edriss S. Titi has been in the University of California, Irvine, since 1988, where he holds a joint appointment in the Department of Mathematics and the Department of Mechanical and Aerospace Engineering.
He received his Ph.D. degree in (1986) from Indiana University. He was a L.E. Dickson Instructor at the University of Chicago (1986-1988), and at the Mathematical Sciences Research Institute in Cornell University (1988-1989). He has been a frequent visitor and consultant to the Center of Nonlinear Studies (CNLS) in the Los Alamos National Laboratory (LANL) since 1989. He was the Orson Anderson Distinguished Visiting Scholar (1997-1998) at the Institute of Geophysics and Planetary Physics and the Stanislaw M. Ulam Scholar (2002-2003) at the CNLS in LANL. In 2004 he was elected Fellow of the Institute of Physics, UK. In 2009 he received the Humboldt Research Award for Senior U.S. Scientists, and also the Society for Industrial and Applied Mathematics (SIAM) Prize on Best Paper in Partial Differential Equations. Since 2011 he is a Fellow of the Center of Smart Interfaces at the Technische Universität Darmstadt, Germany. He was honored in 2012 by being elected as a Fellow of the Society for Industrial and Applied Mathematics (SIAM Fellow).
The research of Edriss S. Titi in applied and computational mathematics lies at the interface between rigorous applied analysis and physical applications. Most of his work has been focused on the development of analytical and computational techniques for investigating nonlinear phenomena. Specifically, in studying the Navier-Stokes equations and other related nonlinear partial differential equations. Such equations arise as models in a wide range of applications in nonlinear science and engineering. The applications include, but are not limited to, fluid mechanics, geophysics, turbulence, chemical reactions, nonlinear fiber optics, and control theory.
Date & Time: Monday, April 9thth, 2012 from 03:00 pm to 04:00 pm
Location: Room # 2418, level 2, building 1 (Al-Khwarizmi)
Refreshments: Available in 3119 @ 02:45 pm
April 09, 2012
The Smile of Mathematicians: Mathematical Modelling and Efficient Simulation in Facial Surgery
By
Dr. Peter Deuflhard (ZIFU & MATHEON)
The talk will give an example from the wide field of virtual medicine, wherein the speaker and his research group have worked during the last years. Here the topic is operation planning in cranio-maxillo facial surgery. In the first part, the establishment of a virtual medical lab is described, essentially the construction of a virtual patient model, both in terms of geometricy and of physiology. The second part will describe efficient algorithms, mostly of adaptive multigrid type, which have been developed for the solution of the linear and nonlinear biomechanics problems. The third part will give illustrations of virtual patients that have been operated. This includes a comparison of the computational predictions with the actual appearance of the real patients. The talk will end with a simulated smile.
Biography: Peter Deuflhard founded and initially presided over the Zuse Institute at the Free University of Berlin and the German national mathematics institute MATHEON, founded in 2002, which develops mathematics for key technologies and supports partners in industry and science. He earned his first degree in physics and his doctorate in Mathematics, studying under R. Bulirsch. He was Professor at the University of Heidelberg before joining the Free University of Berlin in 1986. Professor Deuflhard is an expert in the numerical analysis of fundamental techniques such as linear and nonlinear solvers and initial-boundary value problems and the author of many books on these topics. Throughout the last couple of decades, Peter has made significant mathematical contributions to medicine. His numerous honors include the Maxwell Prize from ICIAM, Fellowship in SIAM, and Membership in the European Academy of Sciences.
Date & Time: Monday, April 9, 2012 from 01:00 to 02:00 pm
Location: Auditorium, level 0, between buildings 4 & 5
Refreshments: Available @ 12:45 pm
April 08, 2012
An Intrusive Polynomial Chaos Method for 3d Navier-Stokes
By
Dr. Chris Lacor (Vrije Universiteit Brussel, Belgium)
First the polynomial chaos method (PCM) is introduced and its advantages to other non-deterministic methods are shortly discussed. An overview of the different implementations of PCM, intrusive and non-intrusive is given. Next, the mathematical formulation of intrusive PCM is discussed and the applications to the 3D compressible Navier-Stokes equations. Simplifying approaches to reduce the CPU cost are presented such as a pseudo spectral approach, truncation of higher order terms and uncoupling of time derivative terms. The method is illustrated with 1D supersonic nozzle with fixed and uncertain geometry, 2D lid driven cavity flow, turbulent RAE2822 airfoil and NASA Rotor 37.
Biography: Chris Lacor got a MSc in Electromechanical Engineering at VUB in 1979. In 1986 he got his PhD from the same university with Charles Hirsch as supervisor. In 1988 he got a permanent position at VUB in the Fluid Mechanics Department and in 1991 he was appointed as professor. His research interests are in schemes and solvers for CFD, Large Eddy Simulation and its applications in Aeroacoustics, Biological Flows and Combustion. Recently, research in non-deterministic flows was started up within the EU project NODESIM-CFD. An intrusive polynomial chaos methodology was developed for 3D Navier-Stokes and applied to several test cases among which airfoils and NASA Rotor 37. Currently, the treatment of geometrical uncertainties is looked into. Chris Lacor was advisor of 17 PhDs. He has more than 50 publications referenced on the Web of Science and is author of more than 100 publications in proceedings of international conferences. He is a member of the CFD committee of ECCOMAS, a fellow of the Flemish Academy of Sciences and an honorary member of the Belgian National Committee on Theoretical and Applied Mechanics.
Date & Time: Sunday, April 8, 2012 from 03:00 to 4:00 pm
Location: Room # 2418, level 2, area 2, building 1, Al Khwarizmi
Refreshments: Available in 2418 @ 02:45 pm
March 25, 2012
Modelling, Control and Observer design for Multi Cells Inverter: the Switched Dynamical Systems Approach
By
Dr. TADJINE Mohamed (ENP Algiers, Algeria)
This presentation deals with modeling and control of multi cell power converter using the theory of switched dynamical systems. The latter is suitable for systems possessing both continuous and discrete states. In the case of multi cells power inverter, the continuous states are the current and voltages signals, while the discrete states are the binary switches that manage the cells. An automata representing the system in its different modes is developed and an hybrid sliding mode control is discussed. Furthermore, the problem of observing the floating voltages of the inverter is treated. Simulations and real time experiments are carried-out to assess the performances and the robustness of the considered methodology.
Keywords: Multi cell inverter, Dynamical hybrid system, Sliding mode control, interconnected systems, Observer design.
Biography: TADJINE Mohamed received the state engineer degree in Automatic Control in 1990 from the ‘Ecole Nationale Polytchnique’, Algiers, Algeria, the DEA degree in Automatic Control and Manufacturing from ‘Instutut National Polytechnique de Grenoble’, INPG Grenoble, France. In 1994, he received the Doctorat degree from INPG, Grenoble France in control and manufacturing management. Between 1994 and 1997 he was a postdoctoral researcher in ‘Automatic Systems Laboratory’, Amiens France. In 1997 he joined ENP Algiers, Algeria as a Teaching Assistant where he is currently Professor in automatic control department. His main research areas are: Non linear and Robust control, Observer design, Fault detection and Fault tolerant control of electrical machines drives.
Date & Time: Sunday, March 25, 2012 from 03:00 to 4:00 pm
Location: Room # 2418, level 2, area 2, building 1, Al Khwarizmi
Refreshments: Available in 2418 @ 02:45 pm