Die DPG-Frühjahrstagung in Dresden musste abgesagt werden! Lesen Sie mehr ...
Bereiche | Tage | Auswahl | Suche | Aktualisierungen | Downloads | Hilfe
DY: Fachverband Dynamik und Statistische Physik
DY 57: Poster: Statistical Physics; Critical Phenomena; Stochastic Thermodynamics; Extreme Events; Data Analytics
DY 57.7: Poster
Donnerstag, 19. März 2020, 15:00–18:00, P1C
Machine Learning the Anderson Transition — Djénabou Bayo1, •Andreas Honecker1, and Rudolf Römer1,2 — 1Université de Cergy-Pontoise, LPTM (UMR8089 of CNRS) , F-95302 Cergy-Pontoise, France — 2University of Warwick, Coventry, CV4 7AL, United Kingdom
The Anderson metal-insulator transition (MIT) is characterized by a transition from a delocalized to a localized state in presence of high disorder. This phenomenon has been investigated for many years and numerical studies have given valuable insight through the determination of the critical properties of the localization length, for example. Machine Learning (ML) and Deep Learning (DL) techniques are still relatively new methods when applied to physics. Recent work shows that ML/DL techniques allow to detect quantum phase transitions directly from images of computed quantum states. The 3D Anderson model is a good candidate for this kind of analysis because of relatively easy access to its quantum states close to the MIT. Here, we implement ML/DL techniques to identify the MIT and to characterize its universal properties. We employ a standard image classification strategy with a multi-layered convolutional neural network. Common ML/DL libraries such as Keras, TensorFlow and FastAI are used in our implementation. We find that a classification by disorder and a reconstructing of the phase diagram appear possible.