Project: Data-Aware Dimensionality Reduction
Description
Dimensionality reduction is one of the leading visualization tools for analyzing multidimensional data, allowing for studying similarity/dissimilarity relationships in any dataset where a distance function can be defined. Despite its popularity, mapping many features to usually a pair of dimensions (in the visual space) trying to keep distance or neighborhood relationships is an approximation, inevitably incurring errors. While different data characteristics may impact the quality of the produced layout differently, the relationships between such data characteristics (e.g., distance distribution or intrinsic dimensionality) and the errors are not well studied in the literature.
This project aims to study the correlation between data characteristics and the different layout errors, such as stress and trustworthiness, focusing on proposing solutions that take advantage of knowing the data properties to adapt the mapping process to reduce errors on the final layout.
Details
- Student
-
DHDiede van der Hoorn
- Supervisor
-
Fernando Paulovich