SVD based initialization A head start for nonnegative matrix factorization.
 
Asynchronous Computation of PageRank computation in an interactive multithreading environment.

Multidamping simulation framework for link%2Dbased ranking.

Machine Learning and Robust Data Mining.

Topic 10 Parallel Numerical Algorithms.

Linear and Non%2DLinear Dimensional Reduction via Class Representatives for Text Classification.

Jylab A System for Portable Scientific Computing over Distributed Platforms.

Oriented k%2Dwindows A PCA driven clustering method.

Exploring term%2Ddocument matrices from matrix models in text mining
We explore a matrix-space model, that is a natural extension to the vector space model for Information Retrieval. Each document can be represented by a matrix that is based on document extracts (e.g. sentences, paragraphs, sections). We focus on the performance of this model for the specific case in which documents are originally represented as term-by-sentence matrices. We use the singular value decomposition to approximate the term-by-sentence matrices and assemble these results to form the pseudo-&#034;term-document&#034; matrix that forms the basis of a text mining method alternative to traditional VSM and LSI. We investigate the singular values of this matrix and provide experimental evidence suggesting that the method can be particularly effective in terms of accuracy for text collections with multi-topic documents, such as web pages with news.
Asynchronous iterative computations with Web information retrieval structures The PageRank case

Parallel Matrix Algorithms and Applications %28PMAA%2704%29.

Asynchronous Iterative Computations with Web Information Retrieval Structures The PageRank Case.

CLSI A Flexible Approximation Scheme from Clustered Term%2DDocument Matrices.

CLSI A Flexible Approximation Scheme from Clustered Term%2DDocument Matrices.

Parallel Matrix Algorithms and Applications %28PMAA %2702%29.

Parallel Computation of Pseudospectra Using Transfer Functions on a MATLAB%2DMPI Cluster Platform.

Parallel computation of pseudospectra by fast descent.
We propose the Pseudospectrum Descent Method (PsDM), a new parallel method for the computation of pseudospectra. The idea behind the method is to use points from an already existing pseudospectrum level curve @  to generate in parallel the points of a new level curve @ such that &lt; . This process can be continued for several steps to approximate several pseudospectrum level curves lying inside the original curve. We show via theoretical analysis and experimental evidence that PsDM is embarrassingly parallel, like GRID and that it adjusts to the geometric characteristics of the pseudospectrum, in particular it captures disconnected components. Results obtained on a parallel system using MPI validate the theoretical analysis and demonstrate interesting load-balancing issues. 1
Parallel computation of pseudospectra by fast descent.
We propose the Pseudospectrum Descent Method (PsDM), a new parallel method for the computation of pseudospectra. The idea behind the method is to use points from an already existing pseudospectrum level curve @  to generate in parallel the points of a new level curve @ such that &lt; . This process can be continued for several steps to approximate several pseudospectrum level curves lying inside the original curve. We show via theoretical analysis and experimental evidence that PsDM is embarrassingly parallel, like GRID and that it adjusts to the geometric characteristics of the pseudospectrum, in particular it captures disconnected components. Results obtained on a parallel system using MPI validate the theoretical analysis and demonstrate interesting load-balancing issues. 1
Topic 19 Problem Solving Environments.

Towards the effective parallel computation of matrix pseudospectra.
Given a matrix A, the computation of its pseudospectrum   (A) is a far more expensive task than the computation of characteristics such as the condition number and the matrix spectrum. As research of the last 15 years has shown, however, the matrix pseudospectrum provides valuable information that is not included in other indicators. So, ask how to compute it eciently and build a tool that would facilitate engineers and scientists to make such analyses? In this paper we focus on parallel algorithms for computing pseudospectra. The most widely used algorithm for computing pseudospectra is embarassingly parallel; nevertheless, it is extremely costly and one cannot hope to achieve absolute high performance with it. We describe algorithms that have drastically improved performance while maintaining a high degree of large grain parallelism. We evaluate the eectiveness of these methods in the context of a MATLAB-based environment for parallel programming using MPI on small, o-the-shelf parallel systems.  Keywords  Pseudospectra, MPI, MATLAB, NOWs  1. 
Cobra Parallel path following for computing the matrix pseudospectrum.
The construction of an accurate approximation of the ffl-pseudospectrum of a matrix by means of the standard grid method is a demanding computational task, even for matrices of medium size. In this paper we describe Cobra, a new domain-based method for the computation of pseudospectra that is a hybrid of the path following based algorithm of Bruhl and the standard grid approach. We implement Cobra using standard LAPACK components and show that it offers large and medium grain parallelism and that it is faster and more robust than the original path following algorithm. The algorithm is also combined with a partial SVD algorithm to produce an effective parallel method for computing the matrix pseudospectrum.
3%2DD Land Avoidance and Load Balancing in Regional Ocean Simulation.

FALCON A MATLAB Interactive Restructuring Compiler.
. The development of efficient numerical programs and library  routines for high-performance parallel computers is a complex task requiring  not only an understanding of the algorithms to be implemented,  but also detailed knowledge of the target machine and the software environment.  In this paper, we describe a programming environment that  can utilize such knowledge for the development of high-performance numerical  programs and libraries. This environment uses an existing highlevel  array language (MATLAB) as source language and performs static,  dynamic, and interactive analysis to generate Fortran 90 programs with  directives for parallelism. It includes capabilities for interactive and automatic  transformations at both the operation-level and the functional- or  algorithm-level. Preliminary experiments, comparing interpreted MATLAB  programs with their compiled versions, show that compiled programs  can perform up to 48 times faster on a serial machine, and up to  140 times fas...
A Parallel Iterative Method for Exponential Propagation.

Workshop on Problem%2DSolving Environments Findings and Recommendations.

Accuracy Control for Parallel Evaluation of Matrix Rational Functions.
Frequently, one needs to evaluate expressions of the form [p(A)]  \Gamma1  q(A)b, where  A 2 R  N \ThetaN  , b 2 R  N  and p; q are polynomialswith degree q  degree p. Algorithms based on the partial fraction representation of q=p when evaluating [p(A)]  \Gamma1  q(A)b  lend themselves to parallelism but may yield poor accuracy. We discuss how to determine an incomplete partial fraction representation of q=p to achieve accuracy and examine its effects on the parallelism. 1 Statement of the Problem and Discussion  In this note we are concerned with the problem of accuracy control in the evaluation of  x := [p(A)]  \Gamma1  q(A)b (1) where A 2 R  N \ThetaN  , b 2 R  N  , p and q are polynomials such that: 1. The degree of q is not larger than the degree of p;  2. Polynomial p has a factorization of the form  Q n j=1  (t \Gamma t j ) with distinct roots  t j ; 3. p and q have no common roots. The above conditions will be assumed to hold throughout this paper. When q j 1, the standard way ...
Experiments with an ocean circulation model on CEDAR.
We present the design of the GFDL ocean circulation model as adapted for simulations of the Mediterranean basin for the Cedar multicluster architecture. The model simulates the basic aspects of large-scale, baroclinic ocean circulation, including treatment of irregular bottom topography. The data and computational mapping strategies and their effect on the design are discussed. The code was parameterized to offer several choices for data partitionings of the computational domain, for placement strategies for the data in the memory hierarchy, and for the number of clusters and processors used in the computational hierarchy of Cedar. The experiments and performance trends are discussed. Using four clusters and 32 processors the code demonstrates significant speedup compared to a single cluster and compared to a single processor.  1 Introduction  The numerical modeling of ocean circulation is a task of great importance, both in its own right, as well as a component of climate studies. Num...
Parallel Ocean Circulation Modeling on Cedar.
The simulation of ocean circulation is a problem of great importance in the environmental sciences of today. Realistic simulations impose great demands on the existing computer systems, requiring the use of considerable computational power and storage capabilities. An implementation of an ocean general circulation model on the Cedar multicluster architecture is presented. This is based on the GFDL three-dimensional model, that was adapted for simulation of the Mediterranean. The model simulates the basic aspects of large-scale, baroclinic ocean circulation, including treatment of irregular bottom topography. The data and computational mapping strategies and the effect on the performance are discussed. The Cedar version of the code, using four clusters and 32 processors, has demonstrated significant speedup compared to a single cluster and compared to a single processor.  iv To my parents, Julio and Niomar De Rose  v ACKNOWLEDGEMENTS This work was supported by the U.S. Department of E...
Experiments with Elliptic Problem Solvers on the Cedar Multicluster.
. The implementation and performance of algorithms for the solution of elliptic problems on the Cedar multiprocessor is examined. The algorithms considered are conjugate gradient schemes with or without preconditioning on a rectangular domain and domain decomposition techniques with overlapping (Schwarz Alternating Method) and nonoverlapping subdomains on a T-shaped domain. We present some preliminary results from our current research effort.  1. Introduction. We have presented elsewhere ([1, 2]) algorithms for solving elliptic PDEs on vector multiprocessors under a memory hierarchy, which in essence consisted of a single cluster of the Cedar architecture designed at CSRD. The major algorithmic design issues in those discussions have been centered around the exploitation of the memory hierarchy and the effective combination of vector and concurrent processing. We are currently modifying and testing algorithms for the Cedar multicluster. In this paper we will provide a brief outline of ...
On the parallel solution of parabolic equations.
We propose new parallel algorithms for the solution of linear parabolic problems. The first of these methods is based on using polynomial approximation to the exponential. It does not require solving any linear systems and is highly parallelizable. The two other methods proposed are based on Pad&#039;e and Chebyshev approximations to the matrix exponential. The parallelization of these methods is achieved by using partial fraction decomposition techniques to solve the resulting systems and thus offers the potential for increased time parallelism in time dependent problems. We also present experimental results from the Alliant FX/8 and the Cray Y-MP/832 vector multiprocessors.  1. Introduction. We consider the following linear parabolic partial differential equation: @u(x; t) @t  = Lu(x; t) + s(x); x  2\Omega  (1.1)  u(0; x) = u 0 ; 8x 2\Omega  u(t; x) = oe(x); x 2  @\Omega ; t  0: where L is a second order partial differential operator of elliptic type, acting on functions defined on the o...
Experiments with Elliptic Problem Solvers on the Cedar Multicluster.
. The implementation and performance of algorithms for the solution of elliptic problems on the Cedar multiprocessor is examined. The algorithms considered are conjugate gradient schemes with or without preconditioning on a rectangular domain and domain decomposition techniques with overlapping (Schwarz Alternating Method) and nonoverlapping subdomains on a T-shaped domain. We present some preliminary results from our current research effort.  1. Introduction. We have presented elsewhere ([1, 2]) algorithms for solving elliptic PDEs on vector multiprocessors under a memory hierarchy, which in essence consisted of a single cluster of the Cedar architecture designed at CSRD. The major algorithmic design issues in those discussions have been centered around the exploitation of the memory hierarchy and the effective combination of vector and concurrent processing. We are currently modifying and testing algorithms for the Cedar multicluster. In this paper we will provide a brief outline of ...


Efficient Parallel Solutions of Parabolic Equations Explicit Methods.

Some Fast Elliptic Solvers on Parallel Architectures and their Complexities.

Fast computation of divided differences and parallel hermite interpolation.
We present parallel algorithms for fast polynomial interpolation. These algorithms can be used for constructing and evaluating polynomials interpolating the function values and its derivatives of arbitrary order (Hermite interpolation). For interpolation, the parallel arithmetic complexity is O(log&amp;sup2; M + log N) for large M and N...
A parallel block cyclic reduction algorithm for the fast solution of elliptic equations.

A Parallel Block Cyclic Reduction Algorithm for the Fast Solution of Elliptic Equations.

Boundary integral domain decomposition of hierarchical memory multiprocessors.
memory multiprocessors
A Parallel Block Cyclic Reduction Algorithm for the Fast Solution of Elliptic Equations.

Fast Laplace Solver by Boundary Integral Based Domain Decomposition Methods.
. A new method called Boundary Integral-based Domain Decomposition is proposed for the fast solution of the Laplace equation on regular and irregular regions. The idea of the method is to adopt the domain decomposition approach by partitioning into subdomains and use an integral equation formulation to compute the interface values. The method provides a particularly efficient algorithm for parallel processing. Preliminary results from numerical experiments on an Alliant FX/8 vector multiprocessor are reported.  1. Introduction. We propose a new method for the fast solution of the Laplace equation on regular and irregular regions. It is called Boundary Integral based Domain Decomposition (BIDD), to encapsulate the ideas behind it: i) partitioning the region into subdomains, ii) computing the interface values by using an integral equation formulation and iii) solving a boundary value problem for each of the subdomains independently. Most of domain decomposition (DD) techniques are based ...
Parallel Rapid Elliptic Solvers.

Numerical Experiments with the Massively parallel Processor.

