What exactly is MLOps, and why does it matter for the business? At Data Summit Connect 2021, Skander Hannachi, AI/ML specialist at Google Cloud, provided an overview of MLOps, how it is used, and why business stakeholders should care.
Hannachi began with an overview of how MLOps has evolved and what it is—a combination of machine learning and development—and then connected how ML relates to the business.
Hannachi demonstrated how MLOps is useful in retail demand forecasting situations with frequently shifting data and in fraud detection, which he noted is even more difficult than retail forecasting scenarios because the subject of the forecasting efforts is trying to evade detection.
With fraud detection, you have data that is shifting far more frequently than in the demand forecasting case because fraudsters are human agents who are intelligent, he said. They sometimes can be pretty advanced machine learning engineers or data scientists themselves, and they will be trying to second guess whatever you're doing to detect fraud and they will be adapting accordingly. If they realize that you have been using the IP address of an incoming transaction as a way to identify whether something was a real transaction they're going to pick up on that. And the moment,their attempted fraud fails, they're going to intelligently adapt.
“This is a situation where MLOps becomes really important because it is essentially impossible to do efficient fraud detection without having a solid MLOps framework in place,” he said.
In addition, Hannachi said, another important aspect of MLOps from a business perspective is being able to explain your predictions. Frequently, there are business stakeholders such as the demand and inventory analysts or security analysts as in the fraud detection case, and some of these business stakeholders will not be technical, yet they still need to be able to understand how a model works.
Moreover, regulatory mandates in fields such as healthcare or real estate make it necessary to have explainable models, not to mention the ethical considerations that make explainable models desirable. “AI is only as good as the humans who train the models,” Hannachi said. “In order to remove bias, we need to be able to understand why the model is making predictions.”
Hannachi’s session was titled “MLOps: A Business Perspective.”
More information about Data Summit Connect 2021 is available here.
Replays of this and all Data Summit Connect 2021 sessions will be available to registered attendees for a limited time and many presenters are making their slide decks available as well.