SAMOP 2023 – wissenschaftliches Programm

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

Q: Fachverband Quantenoptik und Photonik

Q 2: Matter Wave Optics

Q 2.4: Vortrag

Montag, 6. März 2023, 12:00–12:15, E001

Four-Wave Mixing Neurons — •Kai Niklas Hansmann and Reinhold Walser — Institut für Angewandte Physik, Technische Universität Darmstadt, Hochschulstraße 4a, D-64289 Darmstadt, Germany

Artificial neural networks, and especially deep learning, are a rapidly increasing field and have found numerous applications in research and industry over recent years [1]. We propose to implement such a network in a physical system, utilizing the non-linearity of interacting ultracold quantum gases.

For this, we study the four-wave mixing process in bosonic matter waves [2, 3]. Given a superposition of three waves, the four-wave mixing process generates a fourth signal wave. We identify the three initial waves as input and the signal wave as output of an artificial neuron. We show, that the constructed system fulfills all requirements for neuron activity and responds to the input in a non-linear fashion. We perform benchmark calculations to determine the performance of the neuron.

Considering a homogeneous Bose-Einstein condensate in three dimensions with present plane matter waves, we find Josephson-like oscillations beyond the undepleted pump approximation. These can be expressed analytically and agree with numerical Gross-Pitaevskii simulations.

[1] O. Abiodun et al., Heliyon 4, e00938 (2018).

[2] L. Deng et al., Nature 398, 218-220 (1999).

[3] J.M. Vogels et al., Phys. Rev. Lett. 89, 020401 (2002).