Recognizing patterns and structures in data is one of the most popular applications of artificial intelligence and machine learning. Typical application examples range from object recognition in robotics through the early detection of clinical pictures to the autonomous control of vehicles. All of these problems have in common that seemingly large amounts of data are categorized and inspected for similarity to perform targeted tasks via a feedback loop. The images and the data wished to be reconstructed can also be of abstract nature, as they occur in natural scientific experiments as measured variables, or be generated by people themselves, as is the case in literary texts and manuscripts.

The main point of this project is the new development and further development as well as the application of machine learning technologies to the extraction of information from images and data. In recent years, neural networks of various architectures as well as variants, such as e.g. Support vector machines have been widely used to getting impressive results for classification and regression problems. However, considerable efforts are needed to adapt these tools to concrete application fields and make them usable. The performance required for practical applications can only be achieved if structures of the problem area are recognized and exploited. From a theoretical as well as a practical point of view, symmetrical structures, which may also be hidden, are particularly important. These symmetrical structures are a mainstay of the project.

The theoretical foundations and technologies for the use of machine learning to find a quick solution of inversion and recognition problems are purposefully developed for three application fields, which are arranged in three working packages (WP). Optics, Quantum Optics, Nanophotonics, Physical Chemistry (WP1), Plasma Physics (WP2) and Digital Humanities (WP3).