This repository contains the deep semi-supervised learning on CIFAR-10 dataset detecting out of distribution data points (OOD) in unlabeled data in order to improve the performance of the SSL algorithms The repository has the implementation of Two state of the art SSL algorithms: Fixmatch (https://arxiv.org/abs/2001.07685) Unsupervised Data Augmentation (UDA) (https://arxiv.org/abs/1904.12848) This repository uses the SSL impplication if the code in https://github.com/perrying/pytorch-consistency-regularization To run for the FixMatch type
sh ./scripts/fixmatch.sh ./results/fixmatch 2400 for Fixmatch and sh ./scripts/uda.sh ./results/uda 2400 for UDA method (2400) refers to the number of labeled data points abvailable for training
for UDA type
sh ./scripts/uda.sh ./results/uda 2400