Multi/GNN

IBM

Multi/GNN is a tool designed for the detection of money laundering through the use of graph neural networks. The repository includes four classes of models (GIN, GAT, PNA, RGCN) and specific adaptations to identify financial crimes, based on recent research. This resource focuses exclusively on Anti-Money Laundering use cases, and has been developed for experiments in powerful graph neural networks and realistic synthetic financial transactions. It is ideal for public officials and organizations seeking to improve the detection of illicit financial activities.

The Multi-GNN Tool for Anti-Money Laundering solves the problem of financial crime detection by using graph neural network models (GIN, GAT, PNA, RGCN). This solution allows you to identify suspicious transactions and money laundering patterns, improving financial surveillance in governments and organizations.

The Multi-GNN for Anti-Money Laundering tool uses four classes of graph neural network models (GIN, GAT, PNA, RGCN) to detect financial crimes. It is specifically designed for the anti-money laundering use case, allowing experiments with graph neural networks on directed multigraphs.

Multi-GNN consumes CSV transactions and defines experimental settings in JSON and YAML files. Relying on these open, human-readable, and widely supported formats allows GNN models to be trained and replicated across different infrastructures without requiring dedicated databases or software.