It is no exaggeration to say that intelligent systems are rapidly changing the world how we know it. Particularly for image data, machine learning (ML) can handle high variability in high-dimensional data, leading to significant breakthroughs. In many commercial applications of ML, such as photo tagging and machine translation, errors can be tolerated. In contrast, applications in which flaws can have disastrous consequences, such as medical imaging and autonomous driving, have not yet been transformed by artificial intelligence.

Much research is focusing on using either machine learning or estimation. The aim of this group is to consider the bigger picture and focus on the more promising approach in the considered scenario. Further, we also consider combinations of learning-based methods and classical estimators to combine the adaptability and versatility of machine learning with the ease of modeling and low data requirements of classical estimators.