Dresden 2020 – wissenschaftliches Programm

Die DPG-Frühjahrstagung in Dresden musste abgesagt werden! Lesen Sie mehr ...

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

AKPIK: Arbeitskreis Physik, moderne Informationstechnologie und Künstliche Intelligenz

AKPIK 1: AKPIK Talks

AKPIK 1.3: Vortrag

Montag, 16. März 2020, 17:15–17:30, HSZ 301

Detecting noise-blurred deterministic signals with neural networks trained with synthetic data — •Lazaro Alonso Silva, Alexander Eisfeld, Sebastian Gemsheim, and Jan Michael Rost — Max-Planck-Institut für Physik komplexer Systeme, Dresden Germany

We investigate the transferability of artificial neural networks trained on synthetic data with characteristic pattern and noise added to real data to extract the underlying pattern. This is useful when the generation of real training data is computationally expensive or scarce. Using synthetic data from an ideal periodic signal with noise added, we analyse how the extraction capability for a given noise level depends on the size of the training data and the architecture and complexity of the network. We find that networks are most robust in recognizing the periodic signal in noisy data if trained with ideal periodic signals including an optimal amount of noise dependent on the size of the training data. Applied to atomic high harmonic spectra, which are naturally noisy, we can extract their periodicity. In addition, the scheme described allows us to estimate the noise level in the high harmonics spectra.