index.md

The dc_federated package

The dc_federated package is a fedeareted learning library that has been developed at Digital Catapult (UK), by the AI/ML team and the engineering team in London. It has been designed to research, experiment with and demo federated learning, and to deploy consortium scale (< 1000 workers) federated learning applications. Please start at docs/index.md for the full documentation.

Some quick facts about the library

• Which machine learning platforms are supported? The core of the library is platform independent and any machine learning platforms (tensorflow, pytorch, sklearn) or combination of platforms may be used with it.

  • The examples currently included in the library are based on pytorch.

• Which Federated Learning algorthims are supported: The library is designed to be used with any algorithm. The current version includes the FedAvg algorithm implemented with pytorch and instructions on how to implement your own algorithms.

• Is the library ready for deployment? The library has been designed to support consortium level (< 1000 workers) deployment.

• Does the library support worker authentication? The library supports public key authentication.

• Does the library support secure communication? The library supports secure communication via SSL certificates.

Additional Information

You can find additional information in the following locations.

• The Library:
  • Getting Started
  • Library Architecture
  • Library Core Features
  • Implementing New Algorithms
  • Worker Authentication
  • Secure Communication
  • Admin Worker Management
• Examples:
  • MNIST Dataset
  • PlantVillage Dataset
  • Using the FedAvg algorithm
  • Multi-Device Test
• Deployment:
  • Deployment Notes
  • Stress Test