Cross-Bank Context Estimation in Federated Anti-Money Laundering

This method enables multi-layer Graph Neural Network computation for AML by approximating remote account states using transactions recorded by both endpoint banks. Testing on AMLworld HI-Small shows that using transaction counts provides robust gains even in highly fragmented, privacy-preserving environments.