Trustworthy AI in Medicine and Other High-Stakes Domains

Machine-learning systems used in medicine and other high-stakes settings must be evaluated on more than predictive accuracy. Reliability, robustness, interpretability, privacy, and the ability to understand the provenance and limitations of data and predictions are also essential.

We investigate how robust machine learning, causal and concept-based explanations, federated learning, and other trustworthy-AI methods can address these requirements. Current work includes anomaly detection in 3D medical imaging, interpretable clinical AI, distributed learning across healthcare environments, and methods that support data provenance and manipulation detection.