Making CentOS Smarter with Machine Learning

The teams at CentOS and Fedora are planning to play with machine learning on the new stack of hardware they have installed for the continuous integration infrastructure. "We don't have any hard and fast plans for how we anticipate doing that yet or what the workflow is going to be, but it is absolutely something that we are considering," Perrin said.

Perrin said that he would very much want to have capabilities that would at least conduct tests against what users are running. They need to know if the tooling works: if an update reintroduced regression or if an update broke the GPU for acceleration. It's going to have a direct impact on CentOS/Fedora users who are using machine learning.

"We have two distinct use cases for AI – one targeting the users of our software and the second around how to help us provide better code," said Perrin.

"If we see clusters of bugs around particular software, there may be the potential for us to train AI to tell us about bugs or outdated packages. We could use a bot or we could use a piece of machine learning code to automatically fix a particular type of bug, increment that builds, re-spin the code, test it again, see if it passes, and then spit the update out where a person doesn't have to touch it. Ordinarily, it is a six-step process for a person to bump the spec file, build the software, test the software, and put the software out. If we could automate that process via some form of machine learning, fantastic."