SuperSFL: Resource-Heterogeneous Federated Split Learning with Weight-Sharing Supernet

This artifact contains the source code, experiment scripts, and setup instructions associated with the paper "SuperSFL: Weight-Sharing Supernet for Federated Split Learning" submitted to SC 2025.

The artifact enables reproduction of the main experimental results, including training accuracy, communication efficiency, and energy savings comparisons between SFL, DFL, and SuperSFL methods.

Datasets (CIFAR-10 and CIFAR-100) are automatically downloaded and partitioned among clients. Experiments can be launched using simple Python scripts provided in the package.

Hardware used for evaluation includes an NVIDIA A100 GPU and Python 3.9 with PyTorch 2.0.1.

All instructions for setup and execution are provided in the included README.md file.