(3) Visualization
[PIs: Breteler, Klein, Kramer, Lippert, Reuter]
In the last few years particular attention was paid to large collections of data as may be derived from measurements in physics, biology or from repeated simulations with different initial conditions. In this context, we will investigate methods and architectures to approach very large datasets and to reveal important relations between the data. A major question is how the results of analytical machine learning methods can be made interpretable and thus made understandable to the human expert. To this end, novel visualization methods have to be developed and employed for accelerated data analysis, hypothesis generation and quality control of the data. This involves mathematical modeling, adapted tools for supervised and unsupervised learning, compression techniques and the design and implementation of novel tools for computer-aided interactive visual analysis. With the help of such visualization techniques the interpretability of otherwise opaque machine learning techniques can indeed be substantially increased. Further challenges comprise the analysis and visualization of data on varying scales and the propagation and visualization of uncertainty. A particular application will be on the development of computational tools that address issues arising from the constantly increasing availability, degree of automatization, resolution, and flexibility of devices for biomedical image acquisition, e.g. for the Rhineland Study.
