Note: The master
branch is constantly in an unstable or even broken state during development. Please use releases instead of the master
branch in order to get a stable version.
Overview
Federated learning involves multiple parties to collaboratively train a machine learning model, therefore it is usually based on a distributed system. KubeteFATE operationalizes federated learning workloads using cloud native technologies such as containers and Kubernetes. KubeFATE enables federated learning tasks to run across public, private and hybrid cloud environments.
FATE (Federated AI Technology Enabler) is an open-source project to provide a secure computing framework to support the federated AI ecosystem. It implements secure computation protocols based on homomorphic encryption and secure multi-party computation (MPC). It supports federated learning architectures and secure computation of various machine learning algorithms, including logistic regression, tree-based algorithms, deep learning and transfer learning.
KubeFATE supports the deployment of FATE via Docker Compose and Kubernetes. We recommend installing a development environment of FATE via Docker Compose, and a production environment via Kubernetes.
Getting Involved
- For any frequently asked questions, you can check in FAQ.
- Please report bugs by submitting issues.
- Submit contributions using pull requests
Project Structure
KubeFATE
|-- docker-deploy
|-- k8s-deploy
docker-deploy
: The principle is to quickly set up an federated learning environment . Docker Compose can deploy FATE components on a single host. By using Docker Compose, FATE can be set up for environments of multiple parties which are collaborating in a federated manner. Please refer to Docker Compose Deployment for more details.
k8s-deploy
: The deployment approach is designed for a production environment. It is designed with flexibility to operate FATE clusters in different environments.
Major features of KubeFATE k8s-deploy
- Provide a single executable binary for initializing and managing FATE clusters.
- Full life cycle management of FATE clusters, including deploying a new FATE cluster, querying an existing FATE cluster, destroying a given FATE cluster, etc.
- Support customized deployment.
- Support an instance of KubeFATE to manage multiple instances of FATE deployments.
- Provide cluster management service with RESTful APIs.
For more details, please refer to Kubernetes Deployment.
Building KubeFATE
To build the binary of KubeFATE (both CLI and KubeFATE service), a Golang development environment is needed.
$ git clone https://github.com/FederatedAI/KubeFATE.git
$ cd KubeFATE/k8s-deploy/
$ make kubefate-without-swag
To build the container image of KubeFATE service, a Docker environment is needed.
$ make docker-build
Specifying an image repository
By default, the installation script pulls the images from Docker Hub during the deployment. A user could also modify the file .env
to specify a local registry (such as Harbor) to pull images from. A local registry can improve the efficiency of the deployment.