Data Type Agnostic Visual Sensitivity Analysis

Abstract

Illustration of the visualization pipeline used for the Discrepancy Dendrogram. The key idea of it is to use cluster diameters as a measure for variation in complex parameter settings and outputs. Given dissimilarity measures for each, we perform hierarchical clustering separately in each space. Distances are normalized by ranking or min-max normalization for comparison. For every cluster obtained through hierarchical clustering, we evaluate the difference in its diameter and visualize that by color.