Dynamic Sampling For Visual Exploration Of Large Dense-Dense Matrices

We present a technique which allows visual exploration of large dense-occupied similarity matrices. It allows the comparison of several dimensions of a multivariate data set. For the visualization, the data are reduced by sampling. The access time to individual elements is an ever increasing problem with increasing matrix size. We examine various database management systems and compare the access times for different problem sizes. The visualization responds to user interaction and allows the focus to specific areas within the data. For this, the data is filtered according to user interests and the visualization is refined with sub-samples of the filtered data. The context is preserved in this process. The focus allows the discovery of relationships that would otherwise remain hidden.