ghani-haider / flower_tutorials Goto Github PK

Introduction to Flower (Part 1)

• Loading the dataset and centralized training in Pytorch
• Starting on FL
  • Implementing Flower Client Class (using flwr.client.NumPyClient)
  • Implementing client_fn for Virtual Client Enginer (Flower VCE)
• Training using FedAvg Strategy
  • Choosing strategy
  • Configuring client resource
  • Implementing client generated metric (loss, accuracy etc) aggregation function
  • Starting the simulation

Use A Federated Learning Strategy (Part 2)

• Load Torchvision CIFAR10 Dataset into partitions
• Model Training & Eval Pytorch Code For Client
• Customizing FLOWER Client
• Strategy Customizations
  • Server-side parameter initialization: initialized model params at server
  • Changing strategy: FedAvg to FedAdagrad etc
  • Server-side evaluation: test/val at server
  • Sending/receiving arbitrary values to/from clients: configure/set client-side params from server side

Building A Custom Strategy From Scratch (Part 3)

• Dataset Load & Partitions
• Model Architecture, Train & Test Functions
• Custom FLOWER Client
• Helper Functions
• Building A Strategy From Scratch (FedCustom)

Customize The FLOWER Client Class (Part 4)

• Dataset Load & Partitions
• Model Architecture, Train & Test Functions
• Revisiting FLOWER NumPy Client
• Moving from NumPyClient to Client
• Custom Serialization & Deserialization
  • Client Side Implementation (in client class)
  • Server Side Implementation (in strategy class)