The code base is clearly documented and easy to follow.
We provide bash file execution points to train all models.

# ===========ImageNet=================
Download and extract ImageNet first. Then,

To train ImageNet LS experiments, run
> bash bash_scripts/imagenet/train_imagenet_teacher.sh

To train ImageNet KD experiments, place the teacher models as shown in bash script and run,
> bash bash_scripts/imagenet/train_imagenet_student.sh



# ===========CUB200-2011=================
Download the dataset and extract it first. Then,

To train CUB200-2011 LS experiments, run
> bash bash_scripts/cub/train_cub_teacher.sh

To train CUB200-2011 KD experiments, place the teacher models as shown in bash script and run,
> bash bash_scripts/cub/train_cub_student.sh


# ========NMT==============
For NMT, we use the code at https://github.com/RayeRen/multilingual-kd-pytorch and follow exact procedure. Hyper-parameters are reported in the Supplementary.



# ========Visualization of penultimate layer representations==============
For visualization:
> python src/visualization/alpha-LS-KD_imagenet_centroids.py


