Our work reveals that base model architectures (such as MLP and Transformer) struggle in periodic modeling, and proposes Fourier Analysis Network (FAN), a novel neural network that effectively addresses periodicity modeling challenges while offering broad applicability similar to MLP.
🎉 FAN has been accepted to NeurIPS'25.
| MLP Layer | FAN layer | |
|---|---|---|
| Formula | ||
| Num of Params | ||
| FLOPs |
|
|
cd Periodicity_Modeling
bash ./run.sh
Detailed implementations are available in .
The data can be automatically downloaded using the Huggingface Datasets load_dataset function in the ./Sentiment_Analysis/get_dataloader.py.
cd Sentiment_Analysis
bash scripts/Trans_with_FAN/train_ours.sh
bash scripts/Trans_with_FAN/test_ours.shYou can obtain data from Google Drive. All the datasets are well pre-processed and can be used easily.
cd Timeseries_Forecasting
bash scripts/Weather_script/Modified_Transformer.sh cd Symbolic_Formula_Representation
python gen_dataset.py
bash run_train_fan.shcd Image_Recognition
bash run_image_recognition.sh@article{dong2024fan,
title={FAN: Fourier Analysis Networks},
author={Yihong Dong and Ge Li and Yongding Tao and Xue Jiang and Kechi Zhang and Jia Li and Jing Su and Jun Zhang and Jingjing Xu},
journal={arXiv preprint arXiv:2410.02675},
year={2024}
}