Workshop
March 31, 2011
8:00 pm


Workshow Theme: Numerical Analysis of Iterative Linear Algebraic Data Mining Techniques


Modern relational datasets have rich topology and network scientists seek topological understanding. Recent research efforts propose linear algebraic techniques to aide in classifying, ranking, and clustering data entities. For a large dataset, iterative methods are employed to approximate solutions of algebraic equations. These solutions are in turn used to make algorithmic decisions, allowing further analysis to focus on small subsets of the large dataset. Numerical analysis that is aimed at iterative solvers in this context is a blooming research area. This workshop focuses on numerical analysis issues associated with linear algebraic approaches to data and graph analysis.

List of topics:

  • solvers and preconditioning in graphs
  • matrix functions for graph analysis
  • spectral partitioning
  • ranking via linear solves
  • low-rank approximation
  • markov chains
  • applications
  • models
  • impact of skewed degree distribution
  • eigenvector properties
  • analysis of randomized algorithms
  • analysis of convergence tolerances and their relationship to data mining quality
  • early stopping criteria
  • tensor analysis

    Rough outline:

    Iterative Methods for Data Analysis
    by Van Emden Henson
    Sunday afternoon tutorial, 3-4 hours (aimed at graduate-level research assistants in numerical analysis). Monday until possibly Tuesday, parallel session on Numerical Analysis of Iterative Linear Algebraic Data Mining Techniques. Talks for abstract submissions that were either self-identified as part of this thematic track, or identified by the committee. Length depends on how many submissions we will receive.