How ashwinrj/federated-learning-pytorch works
This PyTorch research implementation runs conventional baseline training or federated experiments on MNIST, Fashion-MNIST, and CIFAR-10. The federated workflow loads and partitions training data, selects clients each round, trains local models, averages their weights into a global model, and evaluates it; baseline training follows a separate single-model path. Both paths share model definitions and evaluation support. Dataset acquisition and experiment outputs are shown where source or README evidence supports them; source excerpts do not establish every runtime detail.
Main entry points: federated_main.py Federated coordinator · baseline_main.py Baseline trainer · utils.py Dataset loader · update.py Local training
Components 14
Experiment runners
- Baseline trainer
src/baseline_main.py - Federated coordinator
src/federated_main.py - Run options
src/options.py
Data preparation
- Dataset loader
src/utils.py - Client partitioning
src/sampling.py - Dataset files
data
Models and learning
- MLP and CNNs
src/models.py - Local training
src/update.py - Weight averaging
src/utils.py - Model evaluation
src/update.py
Experiment outputs
- Saved results
save
Other components
- Researcher
- Torchvision datasets
- TensorBoard logging
Connections 19
- Researcher → Baseline trainer: runs baseline
README.md - Researcher → Federated coordinator: runs federated
README.md - Baseline trainer → Run options: parses options
src/baseline_main.py - Federated coordinator → Run options: parses options
src/federated_main.py - Baseline trainer → Dataset loader: loads data
src/baseline_main.py - Federated coordinator → Dataset loader: loads data
src/federated_main.py - Dataset loader → Torchvision datasets: loads datasets
src/utils.py - Torchvision datasets → Dataset files: downloads or reads
- Dataset loader → Client partitioning: assigns client data
src/utils.py - Baseline trainer → MLP and CNNs: constructs model
src/baseline_main.py - Federated coordinator → MLP and CNNs: constructs model
src/federated_main.py - Baseline trainer → Model evaluation: tests model
src/baseline_main.py - Federated coordinator → Local training: trains selected clients
src/federated_main.py - Local training → MLP and CNNs: runs model
src/update.py - Federated coordinator → Weight averaging: averages client weights
src/federated_main.py - Federated coordinator → Model evaluation: tests global model
src/federated_main.py - Local training → TensorBoard logging: logs loss
src/update.py - Baseline trainer → Saved results: saves plot
src/baseline_main.py - Federated coordinator → Saved results: saves metrics
src/federated_main.py