Canonical Ltd. on Tuesday released Charmed Kubeflow 1.4, the newest version of its platform for simplifying enterprise artificial intelligence projects.

U.K.-based Canonical is the maker of Ubuntu, one of the most widely used versions of the Linux operating system. Ubuntu is especially popular in the enterprise, where it’s commonly used to power public cloud environments. The operating system is frequently deployed together with Kubernetes. 

A growing number of enterprises are running AI models in their Kubernetes environments to support machine learning initiatives. In 2018, Google LLC released an open-source tool called Kubeflow to simplify the task of running AI software on Kubernetes. Canonical’s newly updated AI platform, Charmed Kubeflow, is a customized version of Google’s Kubeflow designed to be easier to use.

Canonical provides the software under an open-source license. In addition to Kubeflow’s core features, Charmed Kubeflow includes automation code that the company says simplifies a number of day-to-day management tasks. The software can be deployed in the public cloud, as well as on-premises. 

Charmed Kubeflow’s feature set includes capabilities for only running artificial AI models, but also building and training them. As part of the newly announced update, Canonical has added several features that aim to make developing AI models using the software easier.

The first addition is a component that the company describes as a universal training operator. In AI development projects, developers train neural networks on sample data to improve their accuracy and performance. The operator added by Canonical eases this task.

In Kubernetes environments, where Charmed Kubeflow is designed to be used, an operator is a piece of software that helps manage an application. The operator carries out the configuration tasks involved in setting up the application on Kubernetes. Once the initial setup is complete, it automates day-to-day maintenance tasks such as recovering from outages.

Charmed Kubeflow’s newly added universal training operator makes it easier to deploy and manage the applications that AI developers most commonly use to train neural networks. Charmed Kubeflow supports TensorFlow, MXNet, XGBoost and PyTorch.

Another new feature designed to simplify AI developers’ work is an integration with MLFlow, a popular open-source machine learning toolkit. MLFlow allows developers to store all their neural networks in a centralized library for easy access. The toolkit also helps software teams compare different versions of a neural network to identify which is most effective.

MLFlow can double as a tool for AI drift detection. AI drift is the term for situations where a neural network suddenly starts generating less accurate results, which can occur when data that the neural network is tasked with processing changes. Using MLFlow, Charmed Kubeflow 1.4 provides the ability to detect when an AI’s accuracy declines and automatically retrain it to fix the issue.

The third set of enhancements included in Canonical’s update is aimed at companies where multiple workers use Charmed Kubeflow. Canonical has added features that make it easier to configure the software for multi-user deployment scenarios. 

Canonical provides Charmed Kubeflow for free under an open-source license. The company monetizes the software by selling professional services such as technical support and setup assistance. 

Image: Canonical