GitHub - jubueche/Information-Theory-in-Deep-Learning: This is a tensorflow implementation, aiming at using the Mutual Neural Estimator proposed in (MINE, Belghazi et al) to reconstruct the images of Tishby and Zariv's paper, which can be found here: https://arxiv.org/abs/1703.00810 . Please be aware that the argument of 'compression' in the layers as a cause of extracting features is a little misleading since the mutual information is invariant under nonlinear transformations as described here: https://arxiv.org/abs/1810.05728 which was recently published.

Branches Tags

Name		Name	Last commit message	Last commit date
Latest commit History 2 Commits
MINE.ipynb		MINE.ipynb
desktop.ini		desktop.ini
information_bottleneck_MNIST.ipynb		information_bottleneck_MNIST.ipynb

About

This is a tensorflow implementation, aiming at using the Mutual Neural Estimator proposed in (MINE, Belghazi et al) to reconstruct the images of Tishby and Zariv's paper, which can be found here: https://arxiv.org/abs/1703.00810 . Please be aware that the argument of 'compression' in the layers as a cause of extracting features is a little misle…