[WWW 2026] GFMixer: Decoupled Temporal Gradient and Fourier-Aware Attention for Time Series Forecasting (PDF)

📖 Overview

GFMixer is a dual-path decoupled framework for long-term multivariate time series forecasting and tackles two structural issues in frequency-domain modeling: frequency bias and spectral degradation.

🚀 Quick Start

Environment Setup

To set up the environment, install Python 3.8 with Pytorch 1.4.4. Use the following commands for convenience:

conda create -n GFMixer python=3.8
conda activate GFMixer
pip install -r requirements.txt

Dataset Preparation

Download the pre-processed datasets from:

• Google Drive: Download Link
• Baidu Drive: Download Link

Place the downloaded data in the ./dataset folder.

Running Experiments

Run the following scripts for different forecasting tasks:

⚠️ Important Notes:

• Ensure you have downloaded the datasets and placed them in the correct directory
• The default parameters provided in scripts are a good starting point, but you need to adjust them based on your specific dataset and requirements

📁 Project Structure

GFMixer/
├── README.md                 # Project documentation
├── requirements.txt          # Python dependencies
├── run.py                    # Main entry point for training and testing
├── dataset/                  # Dataset directory
│   ├── ETT/                  # ETT datasets
│   ├── Weather/              # Weather dataset
│   ├── Electricity/          # Electricity dataset
│   ├── Traffic/              # Traffic dataset
│   └── ...
└── ...

⚙️ Training

All scripts are located in ./scripts. For instance, to train a model using the ETTh1 dataset, simply run:

bash ./scripts/GFMixer/ETTh1.sh

After training:

• Your trained model will be safely stored in ./checkpoints.
• Numerical results in .npy format can be found in ./results.
• A comprehensive summary of quantitative metrics is accessible in ./result_long_term_forecast.txt.

🙏 Acknowledgement

Special thanks to the following repositories for their invaluable code and datasets:

• PatchTST
• Time-Series-Library
• Times2D
• RoPE
• PDF

📚 Citation

If you find this repo useful, please consider citing our paper as follows:

@inproceedings{Lin2026GFMixer,
author = {Zhang, Lin and Li, Qing and Zhao, Jingmei},
title = {GFMixer: Decoupled Temporal Gradient and Fourier-Aware Attention for Time Series Forecasting},
year = {2026},
publisher = {Association for Computing Machinery},
address = {New York, NY, USA},
doi = {10.1145/3774904.3792345},
booktitle = {Proceedings of the ACM Web Conference 2026},
pages = {7156–7166},
numpages = {11},
location = {United Arab Emirates},
series = {WWW '26}
}

📩 Contact

If you have any questions, please contact 1230202j1001@smail.swufe.edu.cn, superlin3030@gmail.com or submit an issue.