curriculum-dropout-pytorch
curriculum dropout implementation with pytorch
All experiments can be run on cifar100.
def curriculum_p(p, epoch):
gamma = 0.0001
return (1-(1.-p) * np.exp(-gamma*epoch) + p)| model | acc | Curriculum dropout |
|---|---|---|
| resnet50 | 77.39% | False |
| resnet50 | 77.92% | My |
| resnet50(ours) | 78.71 | True |