An easy, silly, DIY Federated Learning framework with many baselines for individual researchers.


Keywords
fl, pytorch
License
MIT
Install
pip install fedbase==0.6.8

Documentation

FedBase

An easy, modularized, DIY Federated Learning framework with many baselines for individual researchers.

Installation

fedbase @ pypi

pip install --upgrade fedbase

Baselines

  1. Centralized training
  2. Local training
  3. FedAvg, Communication-Efficient Learning of Deep Networksfrom Decentralized Data
  4. FedAvg + Finetune
  5. Fedprox, Federated Optimization in Heterogeneous Networks
  6. Ditto, Ditto: Fair and Robust Federated Learning Through Personalization
  7. WeCFL, On the Convergence of Clustered Federated Learning
  8. IFCA, An Efficient Framework for Clustered Federated Learning
  9. FeSEM, Multi-Center Federated Learning
  10. To be continued...

Three steps to achieve FedAvg!

  1. Data partition
  2. Nodes and server simulation
  3. Train and test

Design philosophy

  1. Dataset
    1. Dataset
      1. MNIST
      2. CIFAR-10
      3. Fashion-MNIST
      4. ...
    2. Dataset partition
      1. IID
      2. Non-IID
        1. Dirichlet distribution
        2. N-class
        3. ...
      3. Fake data
      4. ...
  2. Node
    1. Local dataset
    2. Model
    3. Objective
    4. Optimizer
    5. Local update
    6. Test
  3. Server
    1. Model
    2. Aggregate
    3. Distribute
  4. Server & Node
    1. Topology
    2. Client sampling
    3. Exchange message
  5. Baselines
    1. Global
    2. Local
    3. FedAvg
  6. Visualization

How to develop your own FL with fedbase?