BPCPPDYSOE21 – scientific programme

Parts | Days | Selection | Search | Updates | Downloads | Help

DY: Fachverband Dynamik und Statistische Physik

DY 10: Granular Physics 2 - organized by Matthias Sperl (Köln)

DY 10.5: Talk

Monday, March 22, 2021, 12:20–12:40, DYc

Can machine learning help to identify variables of a granular theory? — •Ansgar Kühn, Song-Chuan Zhao, and Matthias Schröter — Max Planck Institute for Dynamics and Self-Organization, Göttingen

Presently, the best theory for predicting the number of contacts in a granular packing is using the local package fraction as its independent variable [1]. In order to go beyond this one-parameter approach, a more detailed description of the local geometry is given in the form of Minkowski tensors of the Voronoi cell. With this data as features, machine learning provides a more accurate prediction of contact numbers than [1]. Feature selection can be used to identify new variables most relevant for the prediction in order to expand the theory.

[1] Song et al. Nature, 453, 629--632 (2008)