The use of AI has improved computer-aided diagnosis systems in many areas. However, the underlying models are often black boxes that allow the user only limited control over the predictions, which can severely affect the user's trust in the underlying AI. This talk addresses (deep-learning) models where the user can interact with the model and decisions are supported by human expert opinions. Based on a Human-in-the-Loop approach, we envision a close collaboration between the user and the AI foundational to decision making. We aim for more accurate predictions based on the integration of human feedback and demonstrate the applicability of such a collaborative system in a medical use case.
invited by Ralf Engbert