Generative Random Fourier Features

Code for the paper End-to-end Kernel Learning via Generative Random Fourier Features accepted by Pattern Recognition. journal, arxiv.

If our work is helpful for your research, please consider citing:

@article{fang2023end,
  title={End-to-end kernel learning via generative random Fourier features},
  author={Fang, Kun and Liu, Fanghui and Huang, Xiaolin and Yang, Jie},
  journal={Pattern Recognition},
  volume={134},
  pages={109057},
  year={2023},
  publisher={Elsevier}
}

Table of Content

• 1. File descriptions
• 2. Train and attack

1. File descriptions

A brief description for the files in this repo:

• model.py definitions of the GRFF model
• modelv.py definitions of the variant of the GRFF model for image data
• data_loader.py scripts on loading the data
• train.sh & train.py scripts on training the GRFF model on synthetic data and real-world benchmark data
• train_attack_mnist.sh & train_mnist.py & attack_mnist.py scripts on training and attacking the GRFF variant on MNIST

2. Train and attack

Generalization

To see the improved generalization performance of the GRFF model on the synthetic data and the real-world benchmark data, run

sh train.sh

Comment or uncomment specific lines in train.sh to run the corresponding experiments.

Adversarial robustness

To see the adversarial robustness of the GRFF model on MNIST, run

sh train_attack_mnist.sh

Detailed settings of the training hyper-parameters can be found in the 2 scripts above.

If u have problems about the code or paper, u could contact me (fanghenshao@sjtu.edu.cn) or raise issues in this repo.