Data Augmentation in Infinitely Wide Neural Networks

Pre-requisite

• JAX

After installing above packages, run following command

pip install -r requirements.txt

Usage

To use the SNNGP

python snngp_inference.py --model fcn --dataset mnist10k \
--select-method random --num-inducing-points 1000

To use the iSNNGP

python isnngp_inference.py --model fcn --dataset mnist10k \
--select-method random --num-inducing-points 500 \
--augment-X <path to augmented patterns> --augment-y <path to augmented labels>

Note that the augmentations should be stored as .npy files in the 28 x 28 patterns and raw labels.