Recent advances in AI have led to major breakthroughs in many scientific fields, including medical imaging and chemical engineering. Much of the current success can be attributed to the development of deep learning techniques, where algorithms learn from large amounts of data. However, this also poses problems. For example, generalisation is limited when there is distribution shift in the training data, and if the set of parameters is huge and non-intuitive, the data is difficult to interpret.
Therefore, in the KDAI Lab, we are going to strengthen the current data-driven AI by integrating fundamental knowledge from applied natural sciences. We will conduct research on knowledge-driven AI, and show its potential in two applied science domains: medical imaging and chemical engineering. At the same time, our research can be used more broadly, because it studies the fundamental methodology for bringing knowledge into all the key components of AI: data collection, algorithm design, user interaction and implementation. We expect that knowledge-driven AI will be more interpretable and reliable than purely data-driven AI, and that it will further stimulate future scientific development.