(function(doc, html, url) {
var widget = doc.createElement("div");
widget.innerHTML = html;
var script = doc.currentScript; // e = a.currentScript;
if (!script) {
var scripts = doc.scripts;
for (var i = 0; i < scripts.length; ++i) {
script = scripts[i];
if (script.src && script.src.indexOf(url) != -1) break;
}
}
script.parentElement.replaceChild(widget, script);
}(document, '
What is it about?
In this study, we have explored possibilities to generate synthetic crystal structures to use as training sets in deep learning. Specifically, to train neural networks for solving new crystal structures from diffraction data. This is crucial since there is not enough of real data available to successfully train a neural network. The results show that there is virtually no synthetic-to-real domain gap, meaning that neural networks trained on our synthetic structures perform very well on real data.
Why is it important?
Developing new methods for solving crystal structures is of a major importance in many research fields ranging from material science to medicine. Currently, new experimental approaches have been actively developed pushing the limits of the structural science. The cost of that is that quite often the measured experimental data is not sufficiently good to obtain the crystal structure models using the well-established methods. New deep-learning structure solving methods could provide a remedy.