Modern medicine gathers large amounts of very heterogeneous data, ranging from genomics information, proteomics data to histological samples and imaging data of brain and body scans. These medical data sources are enhanced by annotations of medical experts, by environmental and life style information. Data science faces the challenge to extract the relevant bits for diagnosis, prognosis and therapy. Often, algorithm engineers have to design and optimize complex data processing pipelines which have to cope with significant uncertainty and noise. We will demonstrate a machine learning framework to design resilient algorithms for problems in neuroscience and in cancer diagnostics.