The July TDWI Virtual Summit has concluded, but on-demand access is available for previously registered attendees through October 1, 2023.

Click the login button below to access all sessions and content.

Join us for an upcoming summit, or check out our full calendar of virtual training opportunities.

By using website you agree to our use of cookies as described in our cookie policy. Learn More

TDWI Virtual Summit

Solving Business Problems with Analytics

July 12–13, 2023

8:30 am - 12:30 pm PT

Bringing Foundation Models into the Modern MLOps Pipeline

July 13, 2023

Prerequisite: None

James G. Kobielus

Senior Director of Research for Data Management


Foundation models are one of the most important innovations in advanced analytics in this century. They are machine-learning (ML) models that have been built and trained on massive unlabeled data sets, have been optimized for a wide range of tasks, and with a little fine-tuning can easily be adapted to address new tasks.

Foundation models are steadily entering analytics professionals’ core toolsets. Already, there are hundreds of foundation models now available, including those that incorporate transformer, diffusion, and other sophisticated neural-network architectures. Many of these are large language models that perform a wide range of natural language processing tasks, such as automated text generation, classification, and translation. Many generative AI applications leverage foundation models to algorithmically output diverse images, video clips, audio and musical compositions, and much more.

As foundation models come into enterprise data science practices, they will drive changes to MLOps platforms, libraries, workflows, team arrangements, and skill sets. Join TDWI’s senior research director James Kobielus as he examines the promise and status of foundation models in enterprise MLOps. The key emerging enterprise MLOps practices that he will discuss in this regard include:

  • Simplification of the MLOps pipeline through the adoption of foundation models, which learn from unlabeled data sets and thereby eliminate or greatly reduce the need to manually describe each item in a massive data set
  • Scaling of MLOps pipelines to enable very large, complex, and resource-intensive foundation models to do their jobs most effectively
  • Adoption of diverse foundation models within enterprise MLOps initiatives to serve more complex use cases
  • Reliance on MLOps model hubs, marketplaces, and exchanges for discovery, sourcing, and sharing of pretrained and trained foundation models for various use cases
  • Customization of pretrained foundation models to accelerate enterprise MLOps journeys
  • Building responsible governance safeguards into the MLOps pipeline to ensure that foundation models don’t amplify biases implicit in their training data sets, introduce inaccurate or toxic outputs, or violate intellectual property rights

Subscribe to Receive Summit Updates via Email