Bonn 2025 – wissenschaftliches Programm
Bereiche | Tage | Auswahl | Suche | Aktualisierungen | Downloads | Hilfe
A: Fachverband Atomphysik
A 1: Atomic Systems in External Fields I
A 1.3: Vortrag
Montag, 10. März 2025, 12:00–12:15, KlHS Mathe
Can Atoms Learn How to Read? — •Maurice Beringuier1,2 and Thomas Pfeifer1,2 — 1Max Planck Institut für Kernphysik, Saupfercheckweg 1, 69117 Heidelberg — 2Universität Heidelberg, Grabengasse 1, 69117 Heidelberg
Motivated by the potential speed gains of using physical systems for computations we investigate the ability of atomic systems to perform machine-learning tasks.
As in the previous work of Pfeifer et al. (2024, New J. Phys. 26 093018), data and tunable weights are introduced to a simulated atom via the spectral phases of time-dependent electric fields. We compare gradient-free optimization methods and the use of differentiable simulators in their effectiveness to train atoms on the textbook task of recognizing handwritten digits.
We analyze the influence of physical parameters such as the amplitude of the electric field and the level structure of the atoms on its performance on the task.
We identify different phases in the parameter landscape, characterized by the (in-)ability of the atom to learn and correlate these phases with measures that quantify vulnerability to overfitting.
Keywords: Quantum Machine Learning; Machine Learning; Reservoir Computing