Mixup-LR

The implementation of "Enhancing Mixup-based Semi-Supervised Learningwith Explicit Lipschitz Regularization" [ICDM 2020].

How to run?

python train.py (for baseline Mixup)

python train.py --ALR (for the proposed Mixup-LR)

Requirements:

1. PyTorch
2. torchvision

(There might be more requirements but shouldn't be difficult to install them using conda.)

Changes within torchvision

In order to use all data, a separate class CIFAR10All is created inside cifar.py of torchvision. The only difference of this class than the regular CIFAR10 is that it's train list also comprises of test_batch beside reguarl data_batch_i.

Credit:

1. https://github.com/YU1ut/MixMatch-pytorch
2. https://github.com/dterjek/adversarial_lipschitz_regularization