lxcnju / mrtf

MrTF

The source code of our works on federated learning:

• Submitted to ECML-PKDD 2023 Journal Track (Data Mining and Knowledge Discovery, DMKD Journal): MrTF: Model Refinery for Transductive Federated Learning.

Content

• Personal Homepage
• Basic Introduction
• Running Tips
• Citation

Personal Homepage

• Homepage

Basic Introduction

• We consider a real-world scenario that a newly-established pilot project needs to make inferences for newly-collected data, but it does not have any labeled data for training.
• We resort to federated learning (FL) and abstract this scene as transductive federated learning (TFL).
• To facilitate TFL, we propose several techniques including stabilized teachers, rectified distillation, and clustered label refinery.
• The proposed Model refinery framework for Transductive Federated learning (MrTF) shows superiorities towards other FL methods on several benchmarks.
• Related Federated Learning codes could be found in our FL repository FedRepo

Environment Dependencies

The code files are written in Python, and the utilized deep learning tool is PyTorch.

• python: 3.7.3
• numpy: 1.21.5
• torch: 1.9.0
• torchvision: 0.10.0
• pillow: 8.3.1

Datasets

We provide several datasets including (if can not download, please copy the links to a new browser window):

• [MNIST]
• [SVHN]
• [CIFAR-10]
• [CIFAR-100]

Running Tips

• python train_fedavg.py: the baseline of FedAvg
• python train_feddf.py: the baseline of FedDF
• python train_mrtf.py: our proposed algorithm for transductive federated learning.

FL algorithms and hyper-parameters could be set in these files.

Citation

• Xin-Chun Li, Yang Yang, De-Chuan Zhan. MrTF: Model Refinery for Transductive Federated Learning.
• [BibTex]