https://github.com/onehuster/mixupe
Codes for "MixupE: Understanding and Improving Mixup from Directional Derivative Perspective" UAI 2023 Oral
https://github.com/onehuster/mixupe
Last synced: 5 months ago
JSON representation
Codes for "MixupE: Understanding and Improving Mixup from Directional Derivative Perspective" UAI 2023 Oral
- Host: GitHub
- URL: https://github.com/onehuster/mixupe
- Owner: oneHuster
- Created: 2023-08-30T03:44:54.000Z (over 1 year ago)
- Default Branch: main
- Last Pushed: 2023-08-30T05:23:44.000Z (over 1 year ago)
- Last Synced: 2024-08-03T23:14:17.804Z (9 months ago)
- Language: Python
- Size: 154 KB
- Stars: 28
- Watchers: 1
- Forks: 0
- Open Issues: 0
-
Metadata Files:
- Readme: README.md
Awesome Lists containing this project
- Awesome-Mixup - [Code
README
# MixupE
Codes for "MixupE: Understanding and Improving Mixup from Directional Derivative Perspective"
UAI 2023 Best Student Paper
## Dataset
For CIFAR-10/100 and SVHN, these datasets can be downloaded automatically with torchvision.
Script for downloading [Tiny-Imgane-Net](https://gist.github.com/moskomule/2e6a9a463f50447beca4e64ab4699ac4).
## Environments
Environment details used for our study.
```
Python: 3.8.5
PyTorch: 1.11.0+cu10.2
Torchvision: 0.12.0+cu10.2
CUDA: 10.2
NumPy: 1.22.3
scipy 1.5.2
```
## Train with MixupE
```
python main.py --dataset --data_dir --root_dir --train mixupe --mixupe_version 3
```
Take the CIFAR-100 for example,
```
python main.py --dataset cifar100 --data_dir ~/data/CIFAR100/ \
--root_dir ./experiments/ --labels_per_class 5000 \
--arch preactresnet50 --epochs 800 --schedule 10 300 600 --gammas 0.01 0.1 0.1 \
--train mixupe --adv_gen none --adv_train none --mixup_alpha 2.0 \
--decay 0.0001 --mixupe_version 3 --mixup_eta 0.001 --learning_rate 0.1 --threshold 1.0
```
where the hyper-parameters in "--gammas" are "initial learning rate", "learning rate decay factors".
## Citation
```
@inproceedings{zou2023mixupe,
title={Mixupe: Understanding and improving mixup from directional derivative perspective},
author={Zou, Yingtian and Verma, Vikas and Mittal, Sarthak and Tang, Wai Hoh and Pham, Hieu and Kannala, Juho and Bengio, Yoshua and Solin, Arno and Kawaguchi, Kenji},
booktitle={Uncertainty in Artificial Intelligence},
pages={2597--2607},
year={2023},
organization={PMLR}
}
```