Scaleout Studio is a comprehensive, vendor agnostic platform for data-private machine learning. Studio integrates leading open source federated learning with MLOps software into a platform designed to succeed with federated learning in the Enterprise.  Our approach is to build on and extend MLSecOps with federated learning. Studio supports both the the third-party trust-provider role as well as model development in federated learning.

Studio helps you to manage a scalable FL network of aggregation servers and lets you govern client connections. To connect local clients you make use of a light-weight client at each data site or device. In addition to enhanced input privacy from FL, Studio integrates leading open source tools for tracking and model serving, and adds role-based authentication and authorization to all deployed services and endpoints, providing a layer of protection against reverse engineering.

Studio is implemented as a highly modular Django application with a plug-in architecture for third-party applications specified via Helm-charts. This means that you can tailor the deployment to your particular needs, develop your own custom modules, and integrate with any toolchain you rely on for your MLOps pipelines. The platform is optimized for self-managed on-premise usage, and deploys to any standard Kubernetes cluster.

In the base-line distribution we have bundled leading open source tools for model development (Jupyter Lab, VSCode), model tracking and serving (MLFlow, TF Serving, Torch Serve) and storage mangement (MinIO, Kubernetes volumes). Studio provides project-based multi-tenancy, authentication and authorization, object level permissions, and access control for all managed applications and endpoints. Together, this provides an integrated end-to-end collaboration environment for federated learning operations. However, the deployment is composable and you can simply opt-out of any of the bundled MLOps tools if you have an existing infrastructure in place.