Seminar: Graph Learning for Anti-Money Laundering: Advancements and Outlook

Speakers: Lourens Touwen and Çağrı Bilgi, Delft University of Technology, the Netherlands

Title: Graph Learning for Anti-Money Laundering: Advancements and Outlook

Abstract: In recent years, Graph Learning, i.e. deep learning methods applied to graph-structured data, has become a popular area of research. Graph Neural Networks (GNNs) have been especially successful, including in the financial industry. In this talk, we will consider the application of Graph Learning methods to Anti-Money Laundering (AML). Specifically, we consider a bank-transaction network as a multigraph, allowing for parallel edges (transactions) between the same pair of nodes (bank accounts). Existing multigraph neural network architectures either preprocess the multigraph by collapsing it into a simple graph or introduce auxiliary edge features that compromise the permutation equivariance property. We introduce MEGA-GNN, which overcomes these limitations by employing a two-stage aggregation process in the message passing layers: first, parallel edges between the same two nodes are aggregated, and then messages from distinct neighbors are combined. Our experiments show that MEGA-GNN significantly outperforms other multigraph neural networks in tasks such as AML and phishing detection. We end our talk with some open challenges that prevent the real-life application of graph learning solutions in AML, namely data availability and privacy, and consider how Federated Graph Learning may offer a solution.