Summary
The use of artificial intelligence (AI) is becoming of paramount importance for scientific discovery. Scientific AI projects rely upon complex collaborative workflows combining diverse (and potentially sensitive) data sources and artifacts, and as such require reproducibility, accountability, and transparency. Alterations of the AI data or models (either at collection time, at training time or at inference time) pose a severe threat to the accuracy and therefore usability of outcomes from scientific AI. In theory, some of these concerns can be mitigated via federated learning which supports co
