Rostock 2019 – scientific programme

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

Q: Fachverband Quantenoptik und Photonik

Q 41: Poster: Quantum Optics and Photonics II

Q 41.4: Poster

Wednesday, March 13, 2019, 16:15–18:15, S Fobau Physik

Using Langevin Dynamics in Artificial Neural Networks to Represent Quantum Spin Systems — •Felix Behrens, Stefanie Czischek, Martin Gärttner, and Thomas Gasenzer — Kirchhoff-Institut für Physik, INF 227, 69120 Heidelberg, Germany

The idea of connecting artificial neural networks and quantum mechanics gained a lot of interest over the last years. A representation of quantum spin-1/2 states using a specific kind of artificial neural network, the restricted Boltzmann machine, has been introduced by G. Carleo et al. (Science 355, 2017). With an unsupervised learning approach, ground states and dynamics in the system can be found. We implement this ansatz and point out its limitations in the vicinity of a quantum phase transition in the transverse-field Ising model. By varying the setup of the artificial neural network, we find a more flexible representation of quantum many-body systems which can be extended to deeper networks and provides measurements in different spin-bases. Using Langevin-like dynamics, we bring our artificial neural network into a spiking-neural-network-form, which can be implemented on a neuromorphic hardware such as the BrainScaleS system. From this hardware implementation we expect a speedup in the simulations, which offers the possibility of efficiently simulating quantum spin-1/2 systems.