Neural networks learn combinatorial rules.

In our latest PRL we show that combinatorial metamaterials—think of complicated puzzle pieces fitting together to make desired shapes—can be classified efficiently and accurately by neural networks. This is surprising, since these desired shapes occur only for very few specific designs and are sensitive to even a single puzzle piece changing.



Non-Hermitian topology

We discover a form of bulk-edge correspondence in one-dimensional non-Hermitian systems. This applies to non-reciprocal quantum atomic chains and non-reciprocal active mechanical metamaterials. Out in PNAS!Corentin_draftvlowres

Credits illustrations:

Blog at

Up ↑