Accelerated GNN training with DGL and RAPIDS cuGraph in a Fraud Detection Workflow Brad Rees NVIDIA Wesley Chapel, FL, USA brees@nvidia.com Xiaoyun Wang NVIDIA Santa Clara, CA, USA xiaoyunw@nvidia.com Joe Eaton NVIDIA Austin, TX, USA featon@nvidia.com Onur Yilmaz NVIDIA Santa Clara, CA, USA oyilmaz@nvidia.com Rick Ratzel NVIDIA Austin, TX, USA rratzel@nvidia.com Dominque LaSalle NVIDIA Santa Clara, CA, USA dlasalle@nvidia.com ABSTRACT Graph Neural Networks (GNNs) have gained the interest of industry with Relational Graph Convolutional Networks (R-GCNs) showing promise for fraud detection. Taking existing workfows that lever- age graph features to train a gradient boosted decision tree (GBDT) and replacing the graph features with GNN produced embedding achieves an increase in accuracy. However, recent work has shown that the combination of graph attributes with GNN embeddings provides the biggest lift in accuracy. Whether to use a GNN is half of the picture. Data loading, data cleaning and prep (ETL), and graph processing are critical frst steps before graph features or GNN training can be performed. Moreover, the entire process is interactive, optimizing training and validation, for shorter model delivery cycles. Quicker model updates are the key to staying ahead of evolving fraud techniques. McDonald and Deotte [1] published a BLOG on the importance of being able to iterate quickly in fnding a solution. The RAPIDS [2] suite of open-source software libraries gives the data scientist the freedom to execute end-to-end analytics work- fows on GPUs. The ETL and data loading portion is handled by RAPIDS cuDF, which utilizes a familiar DataFrame API. The GBDT process is handled by RAPIDS cuML that has an implementation of XGBoost and RandomForest. The graph analytic portion is handled by RAPIDS cuGraph. Recently cuGraph announced integration into Deep Graph Library (DGL) [3]. For GNN training, graph sampling can consume up to 80% of the training time. RAPIDS cuGraph sampling algorithms execute 10x to 100x faster than similar CPU versions and scale to support massive size graphs. Join us as we dive into GNNs for fraud detection and as we demonstrate how RAPIDS + DGL drastically reduces training time. We will cover everything from accelerating data load and data prep to accelerated GNN training with cuGraph + DGL. ACM Reference Format: Brad Rees, Xiaoyun Wang, Joe Eaton, Onur Yilmaz, Rick Ratzel, and Dom- inque LaSalle. 2022. Accelerated GNN training with DGL and RAPIDS Permission to make digital or hard copies of part or all of this work for personal or classroom use is granted without fee provided that copies are not made or distributed for proft or commercial advantage and that copies bear this notice and the full citation on the frst page. Copyrights for third-party components of this work must be honored. For all other uses, contact the owner/author(s). KDD ’22, August 14–18, 2022, Washington, DC, USA © 2022 Copyright held by the owner/author(s). ACM ISBN 978-1-4503-9385-0/22/08. https://doi.org/10.1145/3534678.3542603 Figure 1: Evolution of Detection Workfow cuGraph in a Fraud Detection Workfow. In Proceedings of the 28th ACM SIGKDD Conference on Knowledge Discovery and Data Mining (KDD ’22), Au- gust 14–18, 2022, Washington, DC, USA. ACM, New York, NY, USA, 2 pages. https://doi.org/10.1145/3534678.3542603 PRESENTER BIOS Brad Rees ś RAPIDS cuGraph Lead, NVIDIA Brad Rees is a Senior Manager at NVIDIA and lead of the RAPIDS cuGraph team. Brad has been designing, implementing, and support- ing a variety of advanced software and hardware systems within the defense and research communities for over 30 years. Brad spe- cializes in complex analytic systems, primarily using graph analytic techniques for social and cyber network analysis. His technical interests are in HPC, machine learning, deep learning, and graphs. Brad has a Ph.D. in Computer Science from the Florida Institute of Technology. Joe Eaton - Principal System Engineer for Graph and Data Analytics, NVIDIA Joe Eaton is the Principal System Engineer for Graph and Data Analytics at NVIDIA. He works on RAPIDS, dividing time between cuML and cuGRAPH. His interests are general optimization and applications of sparse linear algebra to industrial scale problems. Previously, he was manager for sparse linear algebra CUDA libraries cuSPARSE, cuSOLVER, and nvGRAPH, and managed AmgX, now an open-source package of GPU- accelerated sparse iterative solvers. Joe lives in Austin, Texas, and holds a Ph.D. in computational and 4820