This is the code we used in our paper accepted at ICLR workshop 2018
[Semi-Supervised Learning With GANs: Revisiting Manifold Regularization]
Bruno Lecouat*, Chuan Sheng Foo*, Houssam Zenati, Vijay Ramaseshan Chandrasekhar
The repo supports python 3.5 + tensorflow 1.4
To reproduce our results on SVHN
python train_svhn.py
To reproduce our results on CIFAR-10
python train_cifar.py
Here is a comparison of different models using standard architectures (1000 labels on SVHN, and 4000 labels on CIFAR):
Method | SVHN (% errors) | CIFAR (% errors) |
---|---|---|
CatGAN | - | 19.58 +/- 0.46 |
Ladder Network | - | 20.40 +/- 0.47 |
FM | 8.11 +/- 1.3 | 18.63 +/- 2.32 |
ALI | 7.42 +/- 0.65 | 17.99 +/- 1.62 |
VAT small | 6.83 | 14.87 |
Bad GAN | 4.25 +/- 0.03 | 14.41 +/- 0.30 |
Ours | 4.51 +/- 0.22 | 14.45 +/- 0.21 |