Continual Federated Learning

This repository implements C-FLAG from the paper titled "On the Convergence of Continual Federated Learning Using Incrementally Aggregated Gradients" in PyTorch. The framework supports two main paradigms of continual learning:

1. Task-Incremental Learning (Task-IL): Models are trained incrementally across different tasks, where each task has a separate output head.
2. Class-Incremental Learning (Class-IL): Models learn incrementally while adapting to new classes in a unified output space.

Features

• Fully modular and scalable architecture.
• Implemented using PyTorch.
• Supports easy execution with .sh scripts provided in the corresponding directories.

How to Run

To execute experiments, navigate to the relevant directory (e.g., task_incremental or class_incremental) and run the provided .sh files:

bash run_experiment.sh