How oeallen-umass/ece535-federated-learning-bias-analysis works
This repository is a small federated-learning experiment centered on FashionMNIST and FedAvg. The documented workflow partitions training data among simulated clients, trains local model copies, aggregates their weights, then evaluates the global model and compares client-group bias and accuracy variance. The README also describes TERM, AFL, and visualizations, but the supplied tree contains no dedicated implementation files for those capabilities; those branches are documented scope, not confirmed runtime modules.
Main entry points: fedavg.py FedAvg Training · evaluate.py Accuracy Evaluation · data.py FashionMNIST Data · client.py Simulated Clients
Components 11
Data and Clients
- FashionMNIST Data
src/data.py - Data Partitioning
src/partition.py - Simulated Clients
src/client.py
Federated Learning
- FedAvg Training
src/fedavg.py - Weighted Averaging
src/fedavg.py - TERM and AFL
src
Evaluation and Analysis
- Accuracy Evaluation
src/evaluate.py - Bias and Variance
src - Result Visualizations
src
Model
- FashionCNN
src/model.py
Other components
- Experimenter
Connections 16
- Experimenter → FedAvg Training: starts training
src/fedavg.py - FedAvg Training → FashionMNIST Data: loads datasets
src/fedavg.py - FedAvg Training → Data Partitioning: partitions data
src/fedavg.py - Data Partitioning → FashionMNIST Data: reads labels
src/partition.py - FedAvg Training → FashionCNN: creates model
src/fedavg.py - FedAvg Training → Simulated Clients: trains clients
src/fedavg.py - Simulated Clients → FashionCNN: trains local copy
src/client.py - Simulated Clients → FashionMNIST Data: uses assigned data
src/client.py - FedAvg Training → Weighted Averaging: aggregates weights
src/fedavg.py - Weighted Averaging → FedAvg Training: returns averaged weights
src/fedavg.py - FedAvg Training → Accuracy Evaluation: evaluates global model
src/fedavg.py - Accuracy Evaluation → FashionMNIST Data: uses test data
src/evaluate.py - Accuracy Evaluation → FashionCNN: evaluates predictions
src/evaluate.py - Accuracy Evaluation → Bias and Variance: supports comparison
README.md - Bias and Variance → Result Visualizations: visualizes results
README.md - FedAvg Training → TERM and AFL: compares methods
README.md