LR-XFL: Logical Reasoning-based Explainable Federated Learning

Overview

This is the open-source code for AAAI-24 paper "LR-XFL: Logical Reasoning-based Explainable Federated Learning". This repository contains LR-XFL model implementations and the experimental codes that produced the results documented in the paper.

Directory Structure

• entropy_lens/: Core modules of entropy-based network.

  • models/: Model definitions and implementations.
  • logic/: Logic-related modules
  • nn/: Neural network modules

• experiments/: Scripts and data related to different experimental setups.

  • result_plot.py: Script to plot results.
  • data/: datasets used in experiments.
  • mnist.py: Scripts to run LR-XFL, and the baseline FedAvg-Logic on MNIST(Even/Odd) dataset.
  • cub.py: Scripts to run LR-XFL, and the baseline FedAvg-Logic on CUB dataset.
  • vdem.py: Scripts to run LR-XFL, and the baseline FedAvg-Logic on V-Dem dataset.
  • mimic.py: Scripts to run LR-XFL, and the baseline FedAvg-Logic on MIMIC-II dataset.
  • mnist_tree.py: Scripts to run the baseline distributed decision tree (DDT) on MNIST(Even/Odd) dataset.
  • cub_tree.py: Scripts to run the baseline distributed decision tree (DDT) on CUB dataset.
  • vdem_tree.py: Scripts to run the baseline distributed decision tree (DDT) on V-Dem dataset.
  • mimic_tree.py: Scripts to run the baseline distributed decision tree (DDT) on MIMIC-II dataset.

• results/: Directory to store output and results from the project.

Acknowledgement

The local client model is built based on the entropy-lens (https://github.com/pietrobarbiero/entropy-lens). We hereby greatly thank the authors of entropy-lens for their clear code and novel research.