A PyTorch implementation of VIME.
This work is based on the paper by Jinsung Yoon, Yao Zhang, James Jordon, and Mihaela van der Schaar titled "VIME: Extending the Success of Self- and Semi-supervised Learning to Tabular Domain," presented at Neural Information Processing Systems (NeurIPS) in 2020.
The original codebase for "VIME: Extending the Success of Self- and Semi-supervised Learning to Tabular Domain" was implemented using outdated versions of Keras (2.3.1) and TensorFlow (1.15.0). Our aim is to modernize the implementation by transitioning to a more recent version of PyTorch.
Install these packages using conda:
conda install scikit-learn numpy pandas pytorch torchvision -c pytorch
To test the implementation, run the following command:
python semi_pytorch_mnist_test.py