09-14, 13:30–14:00 (Europe/Amsterdam), Foo (main)
Fraud is a major problem for financial services companies. As fraudsters change tactics, our detection methods need to get smarter. Graph neural networks (GNNs) are a promising model to improve detection performance. Unlike traditional machine learning models or rule-based engines, GNNs can effectively learn from subtle relationships by aggregating neighborhood information in the financial transaction networks. However, it remains a challenge to adopt this new approach in production.
The goal of this talk is to share best practices for building a production ready GNN solution and hopefully spark your interest to apply GNNs to your own use cases.
In this talk, we focus on suspicious account detection for online marketplaces. These platforms allow users to set up shops and sell products with little friction. Unfortunately, this attracts fraudsters who abuse these platforms. We use GNNs to do supervised learning based on accounts previously flagged as fraudulent, so that we can learn from both account properties and the relationship between accounts. However, productionizing GNNs is a big challenge. Addressing this challenge purely using open source packages is the main focus of this talk.
We first give an overview of GNN-based fraud detection. Then we deep dive into utilizing PySpark and GraphFrames to build a transaction graph in a scalable way and convert it to DGL (Deep Graph Library) format. Next we share our experiences of setting up training and inference graphs in different time intervals, and deploying the end-to-end model pipeline in Airflow.
Attendees are required to have a basic understanding of machine learning. In this informative talk, they will gain insights into fraud detection's challenges and learn best practices to productionize GNNs.
No previous knowledge expected
Feng is a senior data scientist at Adyen. He is passionate about solving real business problems using innovative AI/machine learning approaches. He received his Ph.D. from the National University of Singapore.
Senior data scientist in Adyen, working in the Score team focusing on fraud detection.
Having PhD background in computer vision and natural language processing using deep neural networks. Familiar with prediction models, such as regression, classification models, etc., as well as the latest research techniques, such as adversarial learning, neural networks etc. Several years of experience with popular deep learning frameworks.