We present a method to visually assess the stability of deep learned projections. For this, we perturb the high-dimensional data by controlled sequences and visualize the resulting changes in the 2D projection. We apply ...