Ryan van Mastrigt

Determining whether a specific mechanical metamaterial design will deform as intended is easy to compute; identifying which designs do so among more than a billion possibilities is not. Neural networks can dramatically accelerate this classification, even when trained on only a tiny fraction of the design space.

In this paper published in Physical Review Letters, we show that neural networks are surprisingly adapt at mapping the design space, despite having sparse training data near the decision boundary. The original article can be found here, and has been covered by popular media outlets in Physics, Materials Today, phys.org and the University of Amsterdam news.