Partitioned Integrators for Thermodynamic Parameterization Of Neural Networks

Supplement code to our Foundations of Data Science 2019 paper.
See: https://www.aimsciences.org/article/doi/10.3934/fods.2019019
Or for the arXiv version: https://arxiv.org/abs/1908.11843

Run the Jupyter notebook AdLaLa.ipynb to see the gap in test accuracy between AdLaLa and Adam for training a fully connected feed-forward network on a binary classification spiral dataset.