Federated Tabular Data Synthesis A Collaboration between Banks

Purpose and goal:

The overall goal of the project is to demonstrate the possibility for banks to collaborate in building data-driven models. This project specifically aims to establish a federated infrastructure between the project parties to train a generative model for synthesizing tabular data. The project will include both technical and legal challenges as the goal is to work with data of a sensitive nature.

The purpose of the project is thus to demonstrate that the technical and legal aspects are manageable and thus to demonstrate that it is possible to collaborate on sensitive data. Expected results and effects:

The project will contribute to knowledge of how infrastructure for federated learning can be established within banking. Furthermore, the legal challenges surrounding sensitive data associated with federated learning are also to be addressed and, to some extent, circumvented. Consequently, all project parties will also build knowledge about what is required in the establishment of federated structures for learning in a real environment.

The result of the project is expected to inspire follow-up projects in federated learning within e.g. identification of money laundering. Approach and implementation:

The project is divided into six work packages where packages 1 and 2 focus on establishing a federated testbed based on AI Swedens Edgelab. Packages 3 and 4 focus on legal aspects and risks within IT to establish a federated architecture. The last two packages are about establishing the federated solution within the banks.