SKM 2023 – wissenschaftliches Programm

Bereiche | Tage | Auswahl | Suche | Aktualisierungen | Downloads | Hilfe

DY: Fachverband Dynamik und Statistische Physik

DY 17: Machine Learning in Dynamics and Statistical Physics I

DY 17.2: Vortrag

Dienstag, 28. März 2023, 10:15–10:30, ZEU 160

Reservoir Computing using Active Matter Model Systems: A Physics Viewpoint — •Mario U. Gaimann and Miriam Klopotek — Stuttgart Center for Simulation Science (SimTech), Cluster of Excellence EXC 2075, University of Stuttgart, Germany

Spatio-temporal prediction of chaotic systems is a challenging problem that is relevant for many fields (weather, finance, energy, and other dynamic systems). Recurrent neural networks and specifically neuron-based reservoir computing were previously used to approach this problem [1,2]. However, these learning systems are typically treated as black boxes, and do not incorporate reasoning or analysis in terms of physical laws and dynamics. Here we study the non-equilibrium dynamics of simple active matter models serving as reservoir computing substrates [3]. This allows us to determine and interpret the state of our reservoir and relate the learning problem to other generic phenomena in statistical physics. With this knowledge we aim to understand optimal conditions for learning in relation to critical states and physical constraints.
 [1] Tanaka, G. et al. (2019), Neural Networks 115, 100-123.
 [2] Nakajima, K. and Fischer, I. (2021). Reservoir Computing. Springer Singapore.
 [3] Lymburn, T. et al. (2021), Chaos 31(3), 033121.