Problems taking AI to production and how to fix them

Description

By working with a variety of clients across many industries—chemical sciences, healthcare, and oil and gas—Jim Scott has documented a number of major impediments to the successful operationalization of these systems and how to keep them running in a production environment, including problems with data formats and optimization and overtime versioning problems of the data, models, and parameters growing too complex. He explores tooling management issues around notebook applications like Jupyter and workflow management tools to keep track and manage an execution pipeline.

New problems arise when when log output volumes grow quickly and significant volume of data movement begins. Source data moves to the GPU, log data moves back to storage, and then the log data moves to machines to handle the distributed compute to perform postmodel analytics to evaluate the performance characteristics. Networks don’t provide infinite bandwidth and most enterprises do not run extremely high-speed networks.

Moving forward, the problems expand when preparing for production deployment of models and adapting them for real time and not just training and testing. You’ll learn about model deployment and scoring with a canary and decoy model leveraging the rendezvous architecture.