Purifying electron spectra from noisy pulses with machine learning using synthetic Hamilton matrices
(Max Planck Institute for the Physics of Complex Systems, Dresden)
XHQ / E1.173 (European XFEL)
XHQ / E1.173
We construct a fully connected feedforward artificial neural network to extract a purified electron spectrum corresponding to ionization with a Fourier limited light pulse from a noisy spectrum created by a short, noisy pulse.
The network is trained by theoretical spectra obtained from a large number of synthetically generated random Hamilton matrices coupled to short pulses and noise. Therefore, application to a wide variety of problems is possible.
Concrete first examples presented will include helium and H2+ for processes dominated by non-linear few-photon absorption in the XUV, where we demonstrate that indeed, the noise free spectrum can be uncovered with good accuracy.