Deploying State-of-the-Art Machine Learning Algorithms at Scale to the Edge: A Case Study
, Accenture Federal Services
, USPS
We'll review the requirements for a MLOPs or DevOps to deliver repeatable, scalable, managed solutions to the edge. Then we'll walk through a specific use case with a major federal client to successfully deploy a modern machine learning application with the ability to process over 84 million events daily. We'll talk about the journey to building custom state-of-the-art machine learning models using NVIDIA’s Object Detection Tool Kit on client-sensitive data, and then leveraging modern DevOps techniques to deploy those models at scale in a controlled, scalable, and expandable architecture while leveraging NGC containers, TensorRT, Triton Inference Server, FastAPI, and Celery. We'll finish with lessons learned for any organization looking to develop custom machine learning applications.