Code of the article "Learning the gauge symmetry using Machine Learning"

this repository contains the code for both tasks described in the following article.

Classifying Gauge orbits

we describe an example of a deep neural network designed for classifying pairs of Ising system $J$ and $J'$ into being in the same gauge orbit or not. In the jupyter-notebook DCNN_simple.ipynb, all the details are commented. Below is an example of the architecture used for this task.

AutoEncoder

We show in the jupyter-notebook AutoEncoder.ipynb, how deep-convolutional neural network can be use to find a latent representation of the gauge orbits of a 2D Ising model with discret $J_{ij} \pm 1$ interactions. All steps are details using the keras library for a system size $L=5$. In addition, it is shown how the latent representation can be used for classfication.