Banks Unite Against Cross-Institutional Fraud with Federated Learning

Adarsh Naidu, a Senior Solutions Architect at Amazon Web Services, has been working on fraud and dispute systems in banking for over two decades. He points out that traditional methods of detecting cross-institutional fraud are flawed because they rely solely on individual banks' transaction data.

Naidu explains that current systems stop at each bank's 'walls,' unable to share customer records due to privacy laws, secrecy rules, and internal controls. This makes it difficult for institutions to learn from one another and detect emerging fraud patterns.

To address this issue, Naidu suggests using federated learning architecture, which allows banks to train models on their own data locally while sharing only the resulting model updates. This approach is combined with differential privacy techniques that add mathematical noise to updates, ensuring individual customer information remains protected.