HARFed: Federated Learning Experiment Simulator

Heterogeneity, Attacks, and Robustness in Federated Learning

Overview

The Harfed application provides a streamlined interface for setting up and conducting Federated Learning experiments, particularly focusing on:

1. Heterogeneous data distributions: Support for advanced dataset partitioning strategies like Dirichlet and Pathological.
2. Robustness testing: Configuring experiments with malfunctions or adversarial behavior (e.g., targeted or untargeted attacks).
3. Monitoring and visualization: GPU utilization tracking, dataset partition visualizations, accuracy trends, and attack success rates over training rounds.

---

Features

Main Functionalities:

• Dataset Configuration:
  • Selection of datasets and subsets.
  • Customizable partitioning strategies (e.g., Dirichlet, IID, etc.).
• Attack Configuration:
  • Integration of Local Differential Privacy (LDP) settings.
  • Fraction of malicious clients.
• Federated Learning Strategies:
  • Popular FL algorithms like FedAvg, FedMedian, FedProx, etc.
  • Flexible settings for model architectures, client fractions, and training rounds.
• Real-time Monitoring:
  • GPU resource tracking (memory, utilization, temperature).
  • Live experiment logs.
• Results Visualization:
  • Accuracy and attack success rate (ASR) plots over FL rounds.
  • Dataset partition label distribution charts.

---

Usage Instructions

1. Install dependencies: Ensure you have Python 3.10+ and install the required packages using pip:

   pip install -r requirements.txt
   

2. Launch the Streamlit app: Run the following command to start the Streamlit application:

   streamlit run app.py
   

3. Configure the experiment:

   • Use the sidebar to select the dataset, partitioner, and attack configurations.
   • Set Federated Learning strategies, the number of clients, and training rounds as needed.

4. Visualize and monitor:

   • Enable dataset partition visualization for insights into the data distribution.

   • Observe live GPU metrics and logs while the experiment is running.

5. Run the experiment:

   • Click on the "Run Experiment" button to execute the experiment.
   • The app generates logs and results at the specified output directory.

6. View and download results:

   • Check the results section for accuracy and ASR plots.
   • Download visualization graphs in PNG or PDF format.
   • Access raw results and GPU stats as CSV files.

---

Screenshots

1. Dataset Partition Visualization

2. Real-time Monitoring

3. Results Visualization