The MixMatch paper looks super interesting for semi-supervised learning: https://arxiv.org/pdf/1905.02249.pdf
Hoping they release the code soon, although I guess it would be a Tensorflow implementation as the authors are all from Google.
“Reduces error rate by up to 4x on CIFAR10”: https://twitter.com/D_Berthelot_ML/status/1125996671664451584