FedROM

Federated Representation Optimal Matching for Robust Federated Learning with Noisy Labels in Remote Sensing

Abstract

This repository contains the implementation of our paper "Representation Optimal Matching for Robust Federated Learning with Noisy Labels in Remote Sensing". We introduce FedROM, a novel federated learning framework that addresses label noise in distributed remote sensing tasks without requiring auxiliary data or transmitting extra sensitive information.

Installation

Key Requirements

• Python 3.8+
• PyTorch 1.11.0

Install Dependencies

pip install -r requirements.txt

Dataset Preparation

Please download the dataset from this link. The dataset file structure should be:

./dataset
├── FedRS
│   ├── NIID-1
│   ├── NIID-2
│   ├── val_balanced
│   └── val_imbalanced
├── FedRS-5
│   ├── NIID-1
│   ├── NIID-2
│   ├── val_balanced
│   └── val_imbalanced
└── FedRS.7z (extract this .7z file to get all datasets)

Usage

Basic Usage

python FL_train.py --alg fedrom --model resnet18 --dataset RS-5 --noise_rate 0.8 --noise_pattern symmetric

Key Parameters

• --alg: Federated learning algorithm (fedrom, fedavg, fedprox, etc.)
• --model: Model architecture (resnet18, alexnet, etc.)
• --dataset: Dataset (RS-5, RS-15)
• --noise_rate: Label noise rate (0.0-1.0)
• --noise_pattern: Noise pattern (symmetric, asymmetric)

For more configuration options, see args.py.

Project Structure

ROT/
├── algorithms/          # Federated learning algorithms
│   ├── fedrom.py       # FedROM core algorithm
│   ├── fedavg.py       # FedAvg algorithm
│   └── ...
├── utils/              # Utility functions
│   ├── model.py        # Model definitions
│   ├── datasets.py     # Dataset processing
│   └── utils.py        # General utilities
├── FL_train.py         # Main training script
├── args.py             # Parameter configuration
└── requirements.txt    # Dependencies

Results

FedROM demonstrates robust performance across 18 baseline methods on two real-world remote sensing datasets with varying label noise rates and multi-source domains.

Citation

If you use this code in your research, please cite our paper:

@article{fedrom2024,
  title={Representation Optimal Matching for Robust Federated Learning with Noisy Labels in Remote Sensing},
  author={Xuefeng Jiang and Co-authors},
  journal={TBD},
  year={2025}
}

License

This project is licensed under the MIT License - see the LICENSE file for details.

Acknowledgments

We sincerely appreciate the authors of the FedRS datasets. More information can be found at FedRS.

⭐ If this project helps you, please give us a star! ⭐