From saunders at stanford.edu Thu Feb 2 09:27:29 2012 From: saunders at stanford.edu (Michael Saunders) Date: Thu Feb 2 09:28:22 2012 Subject: [BANANA] LA/Opt SCREAM seminar TODAY (Ramsharan Rangarajan) Message-ID: Reminder: (Come early -- refreshments served!) SIAM Stanford Student Chapter SCREAM seminar, and Linear Algebra and Optimization Seminar (CME 510) 4:15pm Thursday Feb 02, 2012 Y2E2 101 (473 Via Ortega) Universal meshes: To mesh or not to mesh? Ramsharan Rangarajan PhD student, ME Dept, Stanford How difficult is it to simulate the melting of a block of ice using finite elements? Quite difficult. Why? The geometry of the ice block changes all the time. So how are evolving geometries handled in finite element calculations? I will describe a convenient yet accurate framework for numerically simulating such moving boundary problems. Forthcoming: Thurs Feb 09 Jihye Choi Importance sampling estimators for large deviations Wed Feb 15 Bala Rajaratnam Regularization of positive definite matrices Thurs Feb 16 Arvind Krishna The Bayesian approach for solving Inverse Problems Thurs Feb 23 Dario Grana Bayesian methods in geophysical inverse problems Thurs Mar 01 Chang-han Rhee Sensitivity Analysis of Markov Chains Thurs Mar 08 Jonghyun Lee Bayesian subsurface imaging using total variation prior Thurs Mar 15 Jim Lambers http://www.stanford.edu/group/siam/scream_2012.html http://icme.stanford.edu/seminars/seminars.php http://campus-map.stanford.edu/index.cfm?ID=04-070 -------------- next part -------------- An HTML attachment was scrubbed... URL: http://csmr.ca.sandia.gov/pipermail/banana/attachments/20120202/35ee1b70/attachment.html From mgu at math.berkeley.edu Fri Feb 3 16:36:21 2012 From: mgu at math.berkeley.edu (Ming Gu) Date: Fri Feb 3 16:38:41 2012 Subject: [BANANA] LAPACK seminar Feb 08/Speaker: Aydin Buluc, LBNL Message-ID: <201202040036.q140aLtx028705@phoenix.math.berkeley.edu> Math 290, Section 25, CS 298, Section 6 Spring 2012 (Matrix Computations and Scientific Computing) We meet WEDNESDAYS 12:10 - 1:00PM in Room 380 Soda Hall, Berkeley campus. The coordinators are Profs. J. Demmel (demmel@cs.berkeley.edu) and M. Gu (mgu@math.berkeley.edu). The program will be a mixture of research talks and tutorials. The tutorials will provide a partial sequel to Math 221. For the schedule and other details about the seminar, please see math.berkeley.edu/~mgu/LAPACKSeminar.htm Date: Feb. 08, 2012 Speaker: Aydin Buluc, LBNL Title: Parallel algorithms for sparse matrix product, indexing, and assignment. Abstract: Generalized sparse matrix-matrix multiplication (or SpGEMM) is a key primitive for many high performance graph algorithms as well as some linear solvers, such as algebraic multigrid. SpGEMM also yields elegant parallel algorithms for commonly used sparse matrix primitives, such as sparse matrix indexing (B=A(v,w)) and sparse matrix assignment (A(v,w) = B) for v,w being arbitrary distributed vectors of indices, provided that the underlying SpGEMM implementation is sufficiently flexible and scalable. We demonstrate that our 2D SpGEMM algorithm, together with our serial hypersparse kernels, are indeed highly flexible, scalable, and memory efficient in the general case. This algorithm is the first to yield increasing speedups for an unbounded number of processors, and our experiments show scaling up to thousands of processors in a variety of test scenarios. Based on joint work with John R. Gilbert. Date: Feb. 15, 2012 Speaker: Prof. Jim Demmel, UCB From saunders at stanford.edu Mon Feb 6 17:07:50 2012 From: saunders at stanford.edu (Michael Saunders) Date: Mon Feb 6 17:08:42 2012 Subject: [BANANA] LA/Opt seminar Thursday Feb 9 (Jihye Choi) Message-ID: SIAM Stanford Student Chapter SCREAM seminar, and Linear Algebra and Optimization Seminar (CME 510) 4:15pm Thursday Feb 09, 2012 (Refreshments at 4pm) Y2E2 101, 473 Via Ortega Importance sampling estimators for large deviations Jihye (Julie) Choi PhD student, EE Dept, Stanford jichoi@stanford.edu Importance sampling (IS) is a widely used simulation method for computing rare event probabilities. Existing literature focuses only on constructing an optimal IS estimator. In practice, constructing such an estimator is not easy, and we will be using a suboptimal estimator in most cases. In this talk we study the small sample behavior of IS estimators under an importance measure that is not necessarily optimal. Forthcoming: Wed Feb 15 Bala Rajaratnam Regularization of positive definite matrices Thurs Feb 16 Arvind Krishna The Bayesian approach for solving Inverse Problems Thurs Feb 23 Dario Grana Bayesian methods in geophysical inverse problems Thurs Mar 01 Chang-han Rhee Sensitivity Analysis of Markov Chains Thurs Mar 08 Jonghyun Lee Bayesian subsurface imaging using total variation prior Thurs Mar 15 Jim Lambers http://www.stanford.edu/group/siam/scream_2012.html http://icme.stanford.edu/seminars/seminars.php http://campus-map.stanford.edu/index.cfm?ID=04-070 -------------- next part -------------- An HTML attachment was scrubbed... URL: http://csmr.ca.sandia.gov/pipermail/banana/attachments/20120206/b4599caf/attachment.html From mgu at math.berkeley.edu Mon Feb 6 21:41:12 2012 From: mgu at math.berkeley.edu (Ming Gu) Date: Mon Feb 6 21:43:30 2012 Subject: [BANANA] Reminder: LAPACK seminar Feb 08/Speaker: Aydin Buluc, LBNL In-Reply-To: <201202040036.q140aLtx028705@phoenix.math.berkeley.edu> References: <201202040036.q140aLtx028705@phoenix.math.berkeley.edu> Message-ID: Math 290, Section 25, CS 298, Section 6 Spring 2012 (Matrix Computations and Scientific Computing) We meet WEDNESDAYS 12:10 - 1:00PM in Room 380 Soda Hall, Berkeley campus. The coordinators are Profs. J. Demmel (demmel@cs.berkeley.edu) and M. Gu (mgu@math.berkeley.edu). The program will be a mixture of research talks and tutorials. The tutorials will provide a partial sequel to Math 221. For the schedule and other details about the seminar, please see math.berkeley.edu/~mgu/LAPACKSeminar.htm Date: Feb. 08, 2012 Speaker: Aydin Buluc, LBNL Title: Parallel algorithms for sparse matrix product, indexing, and assignment. Date: Feb. 15, 2012 Speaker: Prof. Jim Demmel, UCB From saunders at stanford.edu Thu Feb 9 13:31:26 2012 From: saunders at stanford.edu (Michael Saunders) Date: Thu Feb 9 13:32:22 2012 Subject: [BANANA] LA/Opt seminar TODAY (Julie Choi) Message-ID: Reminder: seminar this afternoon: SIAM Stanford Student Chapter SCREAM seminar, and Linear Algebra and Optimization Seminar (CME 510) 4:15pm Thursday Feb 09, 2012 (Refreshments at 4pm) Y2E2 101, 473 Via Ortega Importance sampling estimators for large deviations Jihye (Julie) Choi PhD student, EE Dept, Stanford jichoi@stanford.edu Importance sampling (IS) is a widely used simulation method for computing rare event probabilities. Existing literature focuses only on constructing an optimal IS estimator. In practice, constructing such an estimator is not easy, and we will be using a suboptimal estimator in most cases. In this talk we study the small sample behavior of IS estimators under an importance measure that is not necessarily optimal. Forthcoming: Wed Feb 15 Bala Rajaratnam Regularization of positive definite matrices Thurs Feb 16 Arvind Saibaba Computation challenges in the Bayesian approach to solving inverse problems Thurs Feb 23 Dario Grana Bayesian methods in geophysical inverse problems Thurs Mar 01 Chang-han Rhee Sensitivity Analysis of Markov Chains Thurs Mar 08 Jonghyun Lee Bayesian subsurface imaging using total variation prior Thurs Mar 15 Jim Lambers http://www.stanford.edu/group/siam/scream_2012.html http://icme.stanford.edu/seminars/seminars.php http://campus-map.stanford.edu/index.cfm?ID=04-070 -------------- next part -------------- An HTML attachment was scrubbed... URL: http://csmr.ca.sandia.gov/pipermail/banana/attachments/20120209/bf6bd766/attachment-0001.html From mgu at math.berkeley.edu Thu Feb 9 23:58:15 2012 From: mgu at math.berkeley.edu (Ming Gu) Date: Fri Feb 10 00:00:49 2012 Subject: [BANANA] LAPACK seminar on Feb. 15. (Speaker: Jim Demmel) Message-ID: <201202100758.q1A7wFma019641@phoenix.math.berkeley.edu> Math 290, Section 25, CS 298, Section 6 Spring 2012 (Matrix Computations and Scientific Computing) We meet WEDNESDAYS 12:10 - 1:00PM in Room 380 Soda Hall, Berkeley campus. The coordinators are Profs. J. Demmel (demmel@cs.berkeley.edu) and M. Gu (mgu@math.berkeley.edu). The program will be a mixture of research talks and tutorials. The tutorials will provide a partial sequel to Math 221. For the schedule and other details about the seminar, please see math.berkeley.edu/~mgu/LAPACKSeminar.htm Date: Feb. 15, 2012 Speaker: Jim Demmel Title: New Lower and Upper bounds on Communication and Arithmetic Abstract: We present two new approaches to lower and upper bounds in arithmetic and communication. First, we consider dense and sparse linear algebra algorithms: Under reasonable assumptions intended to describe any algorithms that are 3-nested-loop-like, we use geometric arguments to prove lower bounds on the arithmetic required. These bounds are attained for some but not all sparsity patterns, offering the hope of faster algorithms yet to be discovered. Second, in joint work with Katherine Yelick and Michael Christ, we describe a generalization of our communication lower bounds for 3-nested-loop-like algorithms to any number of nested loops, with any array references that are general affine functions of the loop indices. This lower bound is attained by a new algorithm of Yelick and Edgar Solomonik for the N-Body problem. Date: Feb. 22, 2012 Speaker: Katherine Yelick, UCB and LBNL From mgu at math.berkeley.edu Sat Feb 11 00:53:18 2012 From: mgu at math.berkeley.edu (mgu@math.berkeley.edu) Date: Sat Feb 11 00:59:46 2012 Subject: [BANANA] Reminder: LAPACK seminar on Feb. 15. (Speaker: Jim Demmel) In-Reply-To: <201202100758.q1A7wFma019641@phoenix.math.berkeley.edu> References: <201202100758.q1A7wFma019641@phoenix.math.berkeley.edu> Message-ID: <3470c173b007fa6d83e8cb676c417a4c.squirrel@calmail.berkeley.edu> Math 290, Section 25, CS 298, Section 6 Spring 2012 (Matrix Computations and Scientific Computing) We meet WEDNESDAYS 12:10 - 1:00PM in Room 380 Soda Hall, Berkeley campus. The coordinators are Profs. J. Demmel (demmel@cs.berkeley.edu) and M. Gu (mgu@math.berkeley.edu). The program will be a mixture of research talks and tutorials. The tutorials will provide a partial sequel to Math 221. For the schedule and other details about the seminar, please see math.berkeley.edu/~mgu/LAPACKSeminar.htm Date: Feb. 15, 2012 Speaker: Jim Demmel Title: New Lower and Upper bounds on Communication and Arithmetic Date: Feb. 22, 2012 Speaker: Katherine Yelick, UCB and LBNL From saunders at stanford.edu Mon Feb 13 10:24:31 2012 From: saunders at stanford.edu (Michael Saunders) Date: Mon Feb 13 10:25:24 2012 Subject: [BANANA] LA/Opt SCREAM seminar Wednesday (Bala Rajaratnam) Message-ID: SIAM Stanford Student Chapter SCREAM seminar, and Linear Algebra and Optimization Seminar (CME 510) GUEST SPEAKER -- Note special day and place 4:15pm Wednesday Feb 15, 2012 (Refreshments at 4pm) Y2E2 111, 473 Via Ortega (Opposite normal room) Regularization of positive definite matrices: connections between algebra, graph theory, and statistics Professor Bala Rajaratnam Department of Statistics, Department of Environmental Earth System Science, The Woods Institute for the Environment Stanford University http://www-stat.stanford.edu/~brajarat/ Positive definite (p.d.) matrices arise naturally in many areas of mathematics and also feature extensively in scientific applications, including the earth sciences and biomedical sciences. In modern high-dimensional applications, a common approach to finding sparse p.d. matrices is to threshold their small off-diagonal elements (set them to zero). Sometimes referred to as hard-thresholding, this has the attractive property that the resulting matrices are sparse and thus easier to interpret and work with. It is often required, and thus implicitly assumed, that thresholded matrices retain positive definiteness. We formally investigate the algebraic properties of thresholded p.d. matrices. Some interesting and unexpected results will be presented. If time permits, probabilistic properties of thresholded p.d. matrices and connections to optimization will also be discussed. (The presentation will be based on results from three papers: the first by the speaker and D.Guillot, the second by the speaker, B.Naul, D.Guillot and A.Hero, and the third by the speaker.) Forthcoming: Thurs Feb 16 Arvind K. Saibaba Computational challenges in the Bayesian approach to solving inverse problems Thurs Feb 23 Dario Grana Bayesian methods in geophysical inverse problems Thurs Mar 01 Chang-han Rhee Sensitivity Analysis of Markov Chains Thurs Mar 08 Jonghyun Lee Bayesian subsurface imaging using total variation prior Thurs Mar 15 Jim Lambers http://www.stanford.edu/group/siam/scream_2012.html http://icme.stanford.edu/seminars/seminars.php http://campus-map.stanford.edu/index.cfm?ID=04-070 -------------- next part -------------- An HTML attachment was scrubbed... URL: http://csmr.ca.sandia.gov/pipermail/banana/attachments/20120213/7d60f9bf/attachment.html From saunders at stanford.edu Mon Feb 13 10:25:19 2012 From: saunders at stanford.edu (Michael Saunders) Date: Mon Feb 13 10:25:29 2012 Subject: [BANANA] LA/Opt SCREAM seminar Thursday (Arvind Saibaba) Message-ID: Note: We have two SCREAM seminars this week SIAM Stanford Student Chapter SCREAM seminar, and Linear Algebra and Optimization Seminar (CME 510) 4:15pm Thursday Feb 16, 2012 (Refreshments at 4pm) Y2E2 101, 473 Via Ortega Computational challenges in the Bayesian approach to solving inverse problems Arvind K. Saibaba ICME PhD student, Stanford arvindks@stanford.edu In several inverse problems that arise from geophysical applications, such as identifying the contaminant source from time history of spatially distributed contaminant measurements, the Geostatistical approach with Gaussian priors is prevalent. There are two main computational bottlenecks in the large-scale implementation of this approach: (1) Covariance matrices that arise from finely spaced discretizations on irregular grids can be extremely large and dense. (2) For certain problems with a large number of measurements, the measurement operator is not only dense, but forming it explicitly would require repeated solution of (possibly) time-dependent partial differential equations. Using the Hierarchical matrix approach, we will show how to reduce the storage and computational complexity of matrix-vector products involving covariance matrices to O(N\log N), where N is the number of unknowns that we are solving for. The resulting system of equations is solved using a matrix-free Krylov subspace method. We also propose a preconditioner that serves to cluster the eigenvalues of this system of equations and therefore reduce the number of iterations taken by the iterative solver. Further, we shall provide numerical evidence for the clustering of the eigenvalues and demonstrate its performance on a model large-scale inverse problem on unstructured grids based on contaminant source identification. Forthcoming: Thurs Feb 23 Dario Grana Bayesian methods in geophysical inverse problems Thurs Mar 01 Chang-han Rhee Sensitivity Analysis of Markov Chains Thurs Mar 08 Jonghyun Lee Bayesian subsurface imaging using total variation prior Thurs Mar 15 Jim Lambers http://www.stanford.edu/group/siam/scream_2012.html http://icme.stanford.edu/seminars/seminars.php http://campus-map.stanford.edu/index.cfm?ID=04-070 -------------- next part -------------- An HTML attachment was scrubbed... URL: http://csmr.ca.sandia.gov/pipermail/banana/attachments/20120213/5eb9b876/attachment.html From saunders at stanford.edu Wed Feb 15 10:19:27 2012 From: saunders at stanford.edu (Michael Saunders) Date: Wed Feb 15 10:20:19 2012 Subject: [BANANA] LA/Opt SCREAM seminar TODAY () Message-ID: Reminder: Guest speaker today (special day and place) SIAM Stanford Student Chapter SCREAM seminar, and Linear Algebra and Optimization Seminar (CME 510) 4:15pm Wednesday Feb 15, 2012 (Refreshments at 4pm) Y2E2 111, 473 Via Ortega (Opposite normal room) Professor Bala Rajaratnam Department of Statistics, Department of Environmental Earth System Science, The Woods Institute for the Environment Stanford University brajarat@stanford.edu Regularization of positive definite matrices: connections between algebra, graph theory, and statistics Positive definite (p.d.) matrices arise naturally in many areas of mathematics and also feature extensively in scientific applications, including the earth sciences and biomedical sciences. In modern high-dimensional applications, a common approach to finding sparse p.d. matrices is to threshold their small off-diagonal elements (set them to zero). Sometimes referred to as hard-thresholding, this has the attractive property that the resulting matrices are sparse and thus easier to interpret and work with. It is often required, and thus implicitly assumed, that thresholded matrices retain positive definiteness. We formally investigate the algebraic properties of thresholded p.d. matrices. Some interesting and unexpected results will be presented. If time permits, probabilistic properties of thresholded p.d. matrices and connections to optimization will also be discussed. (The presentation will be based on results from three papers: the first by the speaker and D.Guillot, the second by the speaker, B.Naul, D.Guillot and A.Hero, and the third by the speaker.) Forthcoming: Thurs Feb 16 Arvind K. Saibaba Computational challenges in the Bayesian approach to solving inverse problems Thurs Feb 23 Dario Grana Bayesian methods in geophysical inverse problems Thurs Mar 01 Chang-han Rhee Sensitivity Analysis of Markov Chains Thurs Mar 08 Jonghyun Lee Bayesian subsurface imaging using total variation prior Thurs Mar 15 Jim Lambers http://www.stanford.edu/group/siam/scream_2012.html http://icme.stanford.edu/seminars/seminars.php http://campus-map.stanford.edu/index.cfm?ID=04-070 -------------- next part -------------- An HTML attachment was scrubbed... URL: http://csmr.ca.sandia.gov/pipermail/banana/attachments/20120215/12cad7fd/attachment-0001.html From saunders at stanford.edu Thu Feb 16 09:12:57 2012 From: saunders at stanford.edu (Michael Saunders) Date: Thu Feb 16 09:13:44 2012 Subject: [BANANA] LA/Opt seminar TODAY (Arvind K. Saibaba) Message-ID: Reminder: LA/OPT seminar this afternoon SIAM Stanford Student Chapter SCREAM seminar, and Linear Algebra and Optimization Seminar (CME 510) 4:15pm Thursday Feb 16, 2012 (Refreshments at 4pm) Y2E2 101, 473 Via Ortega Computational challenges in the Bayesian approach to solving inverse problems Arvind K. Saibaba ICME PhD student, Stanford arvindks@stanford.edu In several inverse problems that arise from geophysical applications, such as identifying the contaminant source from time history of spatially distributed contaminant measurements, the Geostatistical approach with Gaussian priors is prevalent. There are two main computational bottlenecks in the large-scale implementation of this approach: (1) Covariance matrices that arise from finely spaced discretizations on irregular grids can be extremely large and dense. (2) For certain problems with a large number of measurements, the measurement operator is not only dense, but forming it explicitly would require repeated solution of (possibly) time-dependent partial differential equations. Using the Hierarchical matrix approach, we will show how to reduce the storage and computational complexity of matrix-vector products involving covariance matrices to O(N\log N), where N is the number of unknowns that we are solving for. The resulting system of equations is solved using a matrix-free Krylov subspace method. We also propose a preconditioner that serves to cluster the eigenvalues of this system of equations and therefore reduce the number of iterations taken by the iterative solver. Further, we shall provide numerical evidence for the clustering of the eigenvalues and demonstrate its performance on a model large-scale inverse problem on unstructured grids based on contaminant source identification. Forthcoming: Thurs Feb 23 Dario Grana Bayesian methods in geophysical inverse problems Thurs Mar 01 Chang-han Rhee Sensitivity Analysis of Markov Chains Thurs Mar 08 Jonghyun Lee Bayesian subsurface imaging using total variation prior Thurs Mar 15 Jim Lambers http://www.stanford.edu/group/siam/scream_2012.html http://icme.stanford.edu/seminars/seminars.php http://campus-map.stanford.edu/index.cfm?ID=04-070 -------------- next part -------------- An HTML attachment was scrubbed... URL: http://csmr.ca.sandia.gov/pipermail/banana/attachments/20120216/cf2804c3/attachment.html From mgu at math.berkeley.edu Sun Feb 19 10:38:31 2012 From: mgu at math.berkeley.edu (mgu@math.berkeley.edu) Date: Sun Feb 19 10:45:05 2012 Subject: [BANANA] LAPACK seminar on Feb. 22. (Speaker: Kathy Yelick) In-Reply-To: <3470c173b007fa6d83e8cb676c417a4c.squirrel@calmail.berkeley.edu> References: <201202100758.q1A7wFma019641@phoenix.math.berkeley.edu> <3470c173b007fa6d83e8cb676c417a4c.squirrel@calmail.berkeley.edu> Message-ID: Math 290, Section 25, CS 298, Section 6 Spring 2012 (Matrix Computations and Scientific Computing) We meet WEDNESDAYS 12:10 - 1:00PM in Room 380 Soda Hall, Berkeley campus. The coordinators are Profs. J. Demmel (demmel@cs.berkeley.edu) and M. Gu (mgu@math.berkeley.edu). The program will be a mixture of research talks and tutorials. The tutorials will provide a partial sequel to Math 221. For the schedule and other details about the seminar, please see math.berkeley.edu/~mgu/LAPACKSeminar.htm Date: Feb. 22, 2012 Speaker: Kathy Yelick, UCB and LBNL Date: Feb. 29, 2012 Speaker: W. Kahan, UCB From saunders at stanford.edu Mon Feb 20 19:00:31 2012 From: saunders at stanford.edu (Michael Saunders) Date: Mon Feb 20 19:01:29 2012 Subject: [BANANA] LA/Opt SCREAM seminar Thursday (Dario Grana) Message-ID: SIAM Stanford Student Chapter SCREAM seminar, and Linear Algebra and Optimization Seminar (CME 510) 4:15pm Thursday Feb 23, 2012 (Refreshments at 4pm) Y2E2 101, 473 Via Ortega Bayesian methods in geophysical inverse problems using Gaussian Mixture models Dario Grana PhD student, Geophysics Dept, Stanford dgrana@stanford.edu The main goal of reservoir modeling is to obtain geophysical models of the rock properties in the subsurface, such as porosity, clay content, and fluid saturation. However these properties cannot be measured directly far away from the wells. The only available data are seismic amplitudes, which provide information about the elastic contrasts in the subsurface. Several methods, deterministic and probabilistic, have been developed in order to estimate reservoir properties from seismic data. We propose a new probabilistic methodology that overcomes the common assumption of Gaussian distribution of rock properties. This methodology consists of a full Bayesian approach based on Gaussian mixture models, i.e. linear combinations of Gaussian distributions. We also present a new geostatistical method to solve linear inverse problems in the multimodal case, by using a sequential approach. This methodology is called Sequential Gaussian Mixture Simulation and it allows us to generate multiple realizations of the posterior probability density function of a Gaussian Mixture linear inverse problem. Forthcoming: Thurs Mar 01 Chang-han Rhee Sensitivity Analysis of Markov Chains Thurs Mar 08 Jonghyun Lee Bayesian subsurface imaging using total variation prior Thurs Mar 15 Jim Lambers http://www.stanford.edu/group/siam/scream_2012.html http://icme.stanford.edu/seminars/seminars.php http://campus-map.stanford.edu/index.cfm?ID=04-070 -------------- next part -------------- An HTML attachment was scrubbed... URL: http://csmr.ca.sandia.gov/pipermail/banana/attachments/20120220/8926917a/attachment.html From saunders at stanford.edu Mon Feb 20 20:14:54 2012 From: saunders at stanford.edu (Michael Saunders) Date: Mon Feb 20 20:15:49 2012 Subject: [BANANA] Fwd: [MSandE Faculty] [MSandE All Faculty] OR Seminars -- Wednesday (February 22) -- Karoly Bezdek In-Reply-To: References: Message-ID: Dear LA/Opt colleagues, There's a seminar of interest this week at Stanford (Wed Feb 22, 4:30) ---------- Forwarded message ---------- From: Farnaz Ronaghi Date: Wed, Feb 15, 2012 at 1:45 PM Subject: [MSandE Faculty] [MSandE All Faculty] OR Seminars -- Wednesday (February 22) -- Karoly Bezdek To: or-seminars@lists.stanford.edu Cc: Karoly Bezdek Please join us next Wednesday for an OR seminar talk by Karoly Bezdek: *On Foams with Convex Cells Each Containing a Unit Ball* Karoly Bezdek Computational & Discrete Geometry Professor Centre for Computational & Discrete Geometry Pure Mathematics University of Calgary Wednesday, February 22, 2011 4:30 - 5:30 PM Y2E2, Room 101 *Abstract:* We raise and investigate the following problem: If the Euclidean 3-space is partitioned into convex cells each containing a unit ball, how should the shapes of the cells be designed to minimize the average surface area of the cells? This problem leads to a strong version of the Kepler conjecture on densest sphere packings as well as to a new relative of Kelvin foam problem. The talk is of a survey-type. --++**==--++**==--++**==--++**==--++**==--++**==--++**== or-seminars mailing list or-seminars@lists.stanford.edu https://mailman.stanford.edu/mailman/listinfo/or-seminars --++**==--++**==--++**==--++**==--++**==--++**==--++**==--++**== msande-allfaculty mailing list msande-allfaculty@lists.stanford.edu https://mailman.stanford.edu/mailman/listinfo/msande-allfaculty --++**==--++**==--++**==--++**==--++**==--++**==--++**==--++**== msande-faculty mailing list msande-faculty@lists.stanford.edu https://mailman.stanford.edu/mailman/listinfo/msande-faculty -------------- next part -------------- An HTML attachment was scrubbed... URL: http://csmr.ca.sandia.gov/pipermail/banana/attachments/20120220/5c85e036/attachment-0001.html From mgu at math.berkeley.edu Tue Feb 21 10:56:49 2012 From: mgu at math.berkeley.edu (Ming Gu) Date: Tue Feb 21 11:01:04 2012 Subject: [BANANA] Reminder: LAPACK seminar on Feb. 22, 2012 (Speaker: Kathy Yelick) Message-ID: <201202211856.q1LIunt3027633@phoenix.math.berkeley.edu> Math 290, Section 25, CS 298, Section 6 Spring 2012 (Matrix Computations and Scientific Computing) We meet WEDNESDAYS 12:10 - 1:00PM in Room 380 Soda Hall, Berkeley campus. The coordinators are Profs. J. Demmel (demmel@cs.berkeley.edu) and M. Gu (mgu@math.berkeley.edu). The program will be a mixture of research talks and tutorials. The tutorials will provide a partial sequel to Math 221. For the schedule and other details about the seminar, please see math.berkeley.edu/~mgu/LAPACKSeminar.htm Date: Feb. 22, 2012 Speaker: Kathy Yelick, UCB and LBNL Title: Generalizing Communication Optimal Algorithms Abstract: Recent work by Demmel et al resulted in new communication-avoiding algorithms for dense linear algebra, as well as communication lower bounds against which the algorithms can be shown optimal. In this talk we deconstruct the matrix multiplication lower bound and algorithm to suggest circumstances under which the ideas may be applied to other algorithms or even automated by a compiler. This results in a number of insights, open questions, and a new communication-optimal algorithm for O(n^2) n-body computations. Date: Feb. 29, 2012 Speaker: W. Kahan, UCB From saunders at stanford.edu Thu Feb 23 09:31:35 2012 From: saunders at stanford.edu (Michael Saunders) Date: Thu Feb 23 09:32:28 2012 Subject: [BANANA] LA/Opt SCREAM seminar TODAY (Dario Grana) Message-ID: Reminder: SIAM Stanford Student Chapter SCREAM seminar, and Linear Algebra and Optimization Seminar (CME 510) 4:15pm Thursday Feb 23, 2012 (Refreshments at 4pm) Y2E2 101, 473 Via Ortega Bayesian methods in geophysical inverse problems using Gaussian Mixture models Dario Grana PhD student, Geophysics Dept, Stanford dgrana@stanford.edu The main goal of reservoir modeling is to obtain geophysical models of the rock properties in the subsurface, such as porosity, clay content, and fluid saturation. However these properties cannot be measured directly far away from the wells. The only available data are seismic amplitudes, which provide information about the elastic contrasts in the subsurface. Several methods, deterministic and probabilistic, have been developed in order to estimate reservoir properties from seismic data. We propose a new probabilistic methodology that overcomes the common assumption of Gaussian distribution of rock properties. This methodology consists of a full Bayesian approach based on Gaussian mixture models, i.e. linear combinations of Gaussian distributions. We also present a new geostatistical method to solve linear inverse problems in the multimodal case, by using a sequential approach. This methodology is called Sequential Gaussian Mixture Simulation and it allows us to generate multiple realizations of the posterior probability density function of a Gaussian Mixture linear inverse problem. Forthcoming: Thurs Mar 01 Chang-han Rhee Sensitivity Analysis of Markov Chains Thurs Mar 08 Jonghyun Lee Bayesian subsurface imaging using total variation prior Thurs Mar 15 Jim Lambers http://www.stanford.edu/group/siam/scream_2012.html http://icme.stanford.edu/seminars/seminars.php http://campus-map.stanford.edu/index.cfm?ID=04-070 -------------- next part -------------- An HTML attachment was scrubbed... URL: http://csmr.ca.sandia.gov/pipermail/banana/attachments/20120223/9f5c96f4/attachment.html From mgu at math.berkeley.edu Mon Feb 27 09:24:09 2012 From: mgu at math.berkeley.edu (mgu@math.berkeley.edu) Date: Mon Feb 27 09:30:46 2012 Subject: [BANANA] LAPACK seminar on Feb. 29, 2012 (Speaker: Prof. W. Kahan) In-Reply-To: <201202211856.q1LIunt3027633@phoenix.math.berkeley.edu> References: <201202211856.q1LIunt3027633@phoenix.math.berkeley.edu> Message-ID: Math 290, Section 25, CS 298, Section 6 Spring 2012 (Matrix Computations and Scientific Computing) We meet WEDNESDAYS 12:10 - 1:00PM in Room 380 Soda Hall, Berkeley campus. The coordinators are Profs. J. Demmel (demmel@cs.berkeley.edu) and M. Gu (mgu@math.berkeley.edu). The program will be a mixture of research talks and tutorials. The tutorials will provide a partial sequel to Math 221. For the schedule and other details about the seminar, please see math.berkeley.edu/~mgu/LAPACKSeminar.htm Date: Feb. 29, 2012 Speaker: Prof. W. Kahan, UCB Title: A half-century's Quest to compute Accurate Singular Values quickly Abstract: Some of the history of singular values, and unexpected difficulties encountered by a program to compute them accurately and quickly, will be explored without becoming bogged down in complexities that undermine confidence in a program's correctness. Date: March 7, 2012 Speaker: Lin Lin, LBNL Title: Adaptive local basis set for Kohn-Sham density functional theory in a discontinuous Galerkin framework From saunders at stanford.edu Mon Feb 27 10:03:42 2012 From: saunders at stanford.edu (Michael Saunders) Date: Mon Feb 27 10:04:35 2012 Subject: [BANANA] LA/Opt SCREAM seminar Thursday (Chang-han Rhee) Message-ID: SIAM Stanford Student Chapter SCREAM seminar, and Linear Algebra and Optimization Seminar (CME 510) 4:15pm Thursday March 1, 2012 (Refreshments at 4pm) Y2E2 101, 473 Via Ortega Chang-han Rhee ICME PhD student, Stanford chrhee@stanford.edu Sensitivity analysis of the stationary distribution of Markov chains: Theory, computation and application Consider an optimization problem where the objective function is a steady-state performance measure of a system that can be modeled by a parametrized family of Markov chains. In many computational settings, optimization algorithms converge faster in the presence of computable derivatives. Since the stationary expectation of a stochastic system is typically hard to compute in closed form, a good statistical estimate of the derivative is necessary. We will present how to compute such derivatives efficiently, and discuss when such a performance measure is smooth in the underlying decision variable. Illustrative examples such as AR(n) processes, GI/G/1 queue with heavy tailed service times, beta walk and SDEs will be discussed. The sensitivity analysis problem presented in this talk has a long history, but our result is the first of its kind that covers all canonical examples discussed above and is easy to verify at the same time. A particularly interesting application is found in information theory. There has been significant interest in studying the entropy rate of Markov channels in the information theory community over the past decade. We will show that the entropy rate of Markov communication channel is differentiable with respect to its transition matrix and input/output distributions under mild regularity conditions. Forthcoming: Thurs Mar 08 Jonghyun Lee Bayesian subsurface imaging using total variation prior Thurs Mar 15 Jim Lambers http://www.stanford.edu/group/siam/scream_2012.html http://icme.stanford.edu/seminars/seminars.php http://campus-map.stanford.edu/index.cfm?ID=04-070 -------------- next part -------------- An HTML attachment was scrubbed... URL: http://csmr.ca.sandia.gov/pipermail/banana/attachments/20120227/4fadc69a/attachment.html From mgu at math.berkeley.edu Mon Feb 27 23:20:33 2012 From: mgu at math.berkeley.edu (mgu@math.berkeley.edu) Date: Mon Feb 27 23:27:17 2012 Subject: [BANANA] Reminder: LAPACK seminar on Feb. 29, 2012 (Speaker: Prof. W. Kahan) In-Reply-To: References: <201202211856.q1LIunt3027633@phoenix.math.berkeley.edu> Message-ID: <82e56a4d86236ceca3912303fd7c6592.squirrel@calmail.berkeley.edu> Math 290, Section 25, CS 298, Section 6 Spring 2012 (Matrix Computations and Scientific Computing) We meet WEDNESDAYS 12:10 - 1:00PM in Room 380 Soda Hall, Berkeley campus. The coordinators are Profs. J. Demmel (demmel@cs.berkeley.edu) and M. Gu (mgu@math.berkeley.edu). The program will be a mixture of research talks and tutorials. The tutorials will provide a partial sequel to Math 221. For the schedule and other details about the seminar, please see math.berkeley.edu/~mgu/LAPACKSeminar.htm Date: Feb. 29, 2012 Speaker: Prof. W. Kahan, UCB Title: A half-century's Quest to compute Accurate Singular Values quickly Date: March 7, 2012 Speaker: Lin Lin, LBNL Title: Adaptive local basis set for Kohn-Sham density functional theory in a discontinuous Galerkin framework