Reimagining Fraud Detection through Collaborative Multi-Party GPU Computing
, BNY Mellon
, Inpher
BNY Mellon is one of the world's largest cross-border payments service providers, processing more than $1 trillion daily. Fighting financial crimes and delivering faster transactions has actively pushed the data science teams to explore innovative ways to detect fraud accurately. One of them is training with more extensive, diverse datasets by collaborating with financial partners. These datasets are, however, locked away from sharing due to security and compliance reasons. With the advent of cloud computing, "bad actors" have the same technology as we do. But one thing that they cannot obtain is the scale of training data and computing that we can, by collaborating with our financial partners. As a result, the data science team at BNY built a novel collaborative fraud detection framework leveraging Secure Multi-Party Computation — a cryptographic technique without revealing partners' data. We'll explore this solution and present some of our initial findings, including up to 20% better predictions for fraudulent transactions and performance improvements available through the use of GPUs, which allow for increased data use and higher model precision.