Semi-Supervised Learning tensorflow2.0

This is an Tensorflow implementation of semi-supervised learning with the following methods: Pseudo-label, Pi_model, VAT, mean_teacher, Mixup, ICT and Mixmatch.

Results

CIFAR-10 (4k)

Methods	original paper	ours
MixMatch	93.76±0.06	93.67
Pi_model		86.55
Pseudo-label		84.9
VAT		88.3
VAT_EM		90.11
Mean_teacher		89.47
Mixup		86.5
ICT		92.92

Prerequisites

pip installs:

numpy>=1.17.2
tensorflow-gpu>=2.0
tensorflow-datasets>=1.2.0

Hyperparameters

Epoch = 1024
lr = 0.002
batch-size = 64

Citations

@misc{berthelot2019mixmatch,
    title={MixMatch: A Holistic Approach to Semi-Supervised Learning},
    author={David Berthelot and Nicholas Carlini and Ian Goodfellow and Nicolas Papernot and Avital Oliver and Colin Raffel},
    year={2019},
    eprint={1905.02249},
    archivePrefix={arXiv},
    primaryClass={cs.LG}
}