Differentially Private Secure Multi-Party Computation for Federated Learning in Financial Applications

Federated Learning enables a population of clients, working with a trusted\nserver, to collaboratively learn a shared machine learning model while keeping\neach client's data within its own local systems. This reduces the risk of\nexposing sensitive data, but it is still possible to reverse engineer\ninformation about a client's private data set from communicated model\nparameters. Most federated learning systems therefore use differential privacy\nto introduce noise to the parameters. This adds uncertainty to any attempt to\nreveal private client data, but also reduces the accuracy of the shared model,\nlimiting the useful scale of privacy-preserving noise. A system can further\nreduce the coordinating server's ability to recover private client information,\nwithout additional accuracy loss, by also including secure multiparty\ncomputation. An approach combining both techniques is especially relevant to\nfinancial firms as it allows new possibilities for collaborative learning\nwithout exposing sensitive client data. This could produce more accurate models\nfor important tasks like optimal trade execution, credit origination, or fraud\ndetection. The key contributions of this paper are: We present a\nprivacy-preserving federated learning protocol to a non-specialist audience,\ndemonstrate it using logistic regression on a real-world credit card fraud data\nset, and evaluate it using an open-source simulation platform which we have\nadapted for the development of federated learning systems.\n