Harmonia is an open source project aiming at developing systems/infrastructures and libraries to ease the adoption of federated learning (abbreviated to FL) for researches and production usage. It is named Harmonia, the Greek goddess of harmony, to reflect the spirit of federated learning; that is, multiple parities collaboratively build a ML model for the common good.
The first release includes Harmonia-operator SDK and differential privacy modules (https://github.com/ailabstw/blurnn). We welcome contributions of new aggregation algorithms, privacy mechanism, datasets, etc. Let's work together to flourish the growth of federated learning.
The design of the Harmonia system is inspired by GitOps. GitOps is centerred around a git repository, which maintains the desired states in the production environment. An automated process makes the production environment match the described state in the repository. In Harmonia, training plans (or simply FL parameters), global models, and local models are kept in git repositories. Updates to these repositories trigger FL system state transitions. These automates the FL training processes. A participant in a federated training is composed of an Operator container and an Application container. An Operator container is in charge of maintaining the FL system states, and communicates with an Application container via gRPC. Local training and aggerator applications are encapsulated in Application containers. This container based architecture enables quick plug-in of existing ML workflows.
- docs
- docs/get-started: A step-by-step example tutorial
- docs/sdk: Detailed sdk document
To build the Harmonia Operator,
$ make allfor harmonia/operator image
See examples/mnist
See docs/get-started