How are you securing your Data/ML workflows?

[davidmirror-ops]

For years, Software Engineering (SE) and Machine Learning/Data Science (ML/DS) remained two generally-disjoint fields. In recent years, the adoption of SE practices in ML/DS has increased, mainly centered around the concept of a "pipeline" or "supply chain", or the set of steps needed to get your source code (input data) into an actionable model output (or software artifact).

How to verify the source of each one of the artifacts produced in the supply chain and provide increased transparency remains a largely unaddressed challenge in the SE field, with new standards and frameworks being developed.

In this context, we'd love to know how you are securing (or planning to secure) your ML/DS workflows and what challenges have you experienced on this matter?

We as a community are in a privileged position to start defining what we consider relevant practices for more resilient ML/DS systems, and lead the conversation with organizations both under and outside the Linux Foundation umbrella, to the greater benefit of the entire industry.

Feel free to comment!