An algorithmic framework for visualising and exploring multidimensional data
To help understand multidimensional data, information visualisation techniques are often applied to take advantage of human visual perception in exposing latent structure. A popular means of presenting such data is via two-dimensional scatterplots where the inter-point proximities reflect some notion of similarity between the entities represented. This can result in potentially interesting structure becoming almost immediately apparent. Traditional algorithms for carrying out this dimension reduction tend to have different strengths and weaknesses in terms of run times and layout quality. However, it has been found that the combination of algorithms can produce hybrid variants that exhibit significantly lower run times while maintaining accurate depictions of high-dimensional structure. The author's initial contribution in the creation of such algorithms led to the design and implementation of a software system (HIVE) for the development and investigation of new hybrid variants and the subsequent analysis of the data they transform. This development was motivated by the fact that there are potentially many hybrid algorithmic combinations to explore and therefore an environment that is conductive to their development, analysis and use is beneficial not only in exploring the data they transform but also in exploring the growing number of visualisation tools that these algorithms beget. This thesis descries three areas of the author's contribution to the field of information visualisation. Firstly, work on hybrid algorithms for dimension reduction is presented and their analysis shows their effectiveness. Secondly, the development of a framework for the creation of tailored hybrid algorithms is illustrated. Thirdly, a system embodying the framework, providing an environment conductive to the development, evaluation and use of the algorithms is described. Case studies are provided to demonstrate how the author and others have used and found value in the system across areas as diverse as environmental science, social science and investigative psychology, where multidimensional data are in abundance.