What neural networks teach us about gels
Colloidal particles can form network-like structures that slowly change in time. The dynamics of such gels varies a lot such that some gels are fluid-like while others are solid. However, the structural differences might be small [Gimperlein, Immink, Schmiedeberg, Soft Matter 20, 3143 (2024)].
In a new article we show that neural networks nevertheless can distinguish different types of gels based on purely structural information. Interestingly, even unsupervised learning procedures succeed to classify the gels. We demonstrate that our approach can also be applied to structures that have been determined in experiments. Therefore, neural networks recognize the subtle structural differences that lead to very diverse dynamical properties [Gimperlein, Schmiedeberg, Eur. Phys. J. E 48, 5 (2025)].