DIMACS Workshop on Algorithms for Multidimensional Scaling

August 10, 2001
DIMACS Center, Rutgers University, Piscataway

Organizers:
J. Douglas Carroll, Rutgers University, dcarroll@rci.rutgers.edu
Phipps Arabie, Rutgers University, arabie@andromeda.rutgers.edu
Lawrence J. Hubert, University of Illinois, lhubert@s.psych.uiuc.edu
Presented under the auspices of the Special Focus on Data Analysis and Mining.

Talks will primarily focus on algorithmic aspects of MDS. Primary emphasis will be given to use of algorithms not typically used for fitting MDS models, such as linear and mixed integer programming, nonlinear or dynamic programming, and other optimization methods that have not been traditionally applied to fitting MDS and related models-- especially when these can be used to fit models not easily amenable to more traditional optimization techniques, such as various gradient based procedures. Another class of algorithmic issues to be considered has to do with approaches for increasing the size of data sets MDS and related methods can deal with, either via improvements in existing algorithms aimed at speeding them up considerably, as well as enabling them to deal with larger data sets, or by use of heuristic methods that may not precisely optimize a well defined criterion of fit, but may allow dealing with much larger data sets efficiently. Papers of this type can generically be classified as papers on MDS and related techniques for Massive Data Sets (MDS for MDS), or "Data Mining" in the context of MDS and related methodology.

The class of MDS and related models that will be dealt with include, in addition to spatial models for proximity data, multidimensional or multiattribute models for preferential choice or other multivariate data, non-spatial or discrete models, such as tree structure, (overlapping or non-overlapping) clustering models, or "hybrid" models combining aspects of continuous spatial and discrete non-spatial models (e.g., a model for proximity data in which proximities are related to a sum of distances from an MDS-like spatial model and an ultrametric or path length metric defined on one or more tree structures; alternatively, the discrete component could consist of distances or distance-like measures defined on pairs of objects based on, say, an overlapping clustering structure).


Next: Call for Participation
Workshop Index
DIMACS Homepage
Contacting the Center
Document last modified on May 9, 2001.