Shake Drop regularization
April 24, 2020 ยท View on GitHub
PyTorch implementation of shake-drop regularization.
Dependencies
- python 3.5+
- PyTorch 1.0.0
Accuracy
CIFAR-100
| Model | Method | Level | Alpha | Beta | This implementaion | Paper |
|---|---|---|---|---|---|---|
| PyramidNet | ShakeDrop | Batch | [-1, 1] | [0, 1] | 83.90 | 83.78 |
Train PyramidNet(depth=110, alpha=270) with shake-drop for CIFAR-100
python train.py --epochs 300 --batch_size 128 --label 100 --lr 0.5 --depth 110 --alpha 270 --snapshot_interval 10
References
Yoshihiro Yamada, Masakazu Iwamura, Koichi Kise. "ShakeDrop regularization" ICLR2018 UnderReview
Yoshihiro Yamada, Masakazu Iwamura, Takuya Akiba, Koichi Kise. " ShakeDrop Regularization for Deep Residual Learning" arXiv:1802.02375v2
Author Torch Implementation https://github.com/imenurok/ShakeDrop