- Talk for All Things Open 2025
Description: Putting together a working ML pipeline can make you feel like Dr. Frankenstein. One day you're digging up fresh data in the local graveyard. The next, you're picking out the right frameworks and versions to bring your monster to life. And if you're not careful, the final product, your model, may turn on you.
In this spookifying talk, Patrick will walk us through the art and science of stitching together a secure ML pipeline in 2025. From choosing and safely loading base models to determining the provenance of components such as containers in your infrastructure, we'll outline how to train dependable models and run them safely in inference. Along the way, we'll prepare for Halloween with hair-raising stories of data poisoning, supply chain attacks, and agency excesses. By the time we're done, you'll be a a modern Prometheus, bringing the forbidden fire of secure model training to your organization.
m