Discussion |
| |
Authors: | John F Elder IV |
| |
Affiliation: | Department of Computational and Applied Mathematics , Rice University , Houston , TX , 77005 , USA |
| |
Abstract: | Abstract Visualization is increasingly being recognized as an effective and efficient way not only to communicate patterns in scientific data, but to discover them as well. In the low dimensions of everyday experience, the human ability to find meaningful order in noisy data may never be matched by automatons. So the introduction of a useful visualization procedure, as provided here by Furnas and Buja, is indeed a welcome development. They show that low-dimensional patterns extracted by a combined projection and section operation (a prosection) can imply the existence of similar higher-dimensional structure. In exploratory or inductive data analyses, then, prosections could be used to generate hypotheses about relationships between sampled variables. However, the familiar curse of dimensionality may confine their practical application to point clouds of only moderate dimension. |
| |
Keywords: | Atypical data Dimensionality reduction Neural networks Nonlinear projections Self-organization |
|
|