Long-Range Order Hidden in Short-Range Disorder: Challenging the Limits of Reliable CNN Classification

Kurzfassung

The talk was a collection of the latest insights of the research conducted on training a basic CNN to predict material properties. It gave an introduction on the creation of hyperuniform, black-and-white point patterns in 2D and its use as training datasets. Further, the process of optimizing and fine-tuning this CNN was illustrated together with some preliminary prediction results.