The goal of this project was to determine if artificial neural networks trained on MNIST digits are poised at criticality.
This code has 3 main functions:
-Create a neural network designed to classify MNIST digits
-Plot the heat capacity of the activations of the neural network and see if it matches the heat capacity plots of known critical systems
-Create energy vs entropy plots to further confirm criticality
Disclaimer: There might be some bugs in the code, I have quickly made a repo of the code because looking at my old google colab doc was making me upset at how bad the code was
This code is organized into a few folders
-figures: figures stores all the nice and fun plots
-sources: The bulk of the code is included in the sources. Sources includes code to calculate heat capacity, entropy, energy, etc.
-scripts: Runs the code written in sources, makes the pretty plots
-tensorflow
-matplotlib
-keras
-numpy
-math
-scipy