Clustering in Machine Learning Pattern Formation of VO2

<p>We have trained a convolutional neural network (CNN) machine learning (ML) model to recognize images from seven different candidate Hamiltonians that could be controlling pattern formation of metal-insulator domains in Vanadium Dioxide (VO<sub>2</sub>).  This trained CNN was then applied to experimental data on VO<sub>2</sub> taken via scanning near-field infrared microscopy and via optical microscopy.  </p>

<p>The html files are interactive 3-dimensional plots of the full 7-dimensional that visualize the output of the last fully connected layer of the CNN from the training sets.  The plots also contain the high and low confidence predictions from experimental data. </p>

<p>The included codes contain the architecture of the ML model, the symmetry reduction operations applied to binary images, and the criterion for assigning high or low confidence to the ML classifier predictions.  <br />
 </p>

<p> </p>