TRR 318-2 - Project C03: Interpretable machine learning—Explaining change

C03 develops methods to explain how and why machine learning models adapt over time. Our project makes AI systems more transparent in dynamic settings by designing efficient, expressive explanations. This includes novel approaches for real-time explanation and higher-order Shapley interactions. In the second funding period, we aim for ex-ante (rather than post-hoc) methods applicable to compound AI-systems (rather than homogeneous AI-models). We build fundamental computational tools that enable the TRR to analyze, communicate, and monitor AI behavior in dynamic human–AI collaboration.