fedbase 0.2.18

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

FedBase

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

Installation

fedbase @ pypi

pip install --upgrade fedbase

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 classes
         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?

Baselines

1. Centralized train
2. Local train
3. FedAvg
4. Ditto
5. Clustered FL
6. ...

To be continued...