Aug 28 – 30, 2024
BC Building, Swiss Federal Institute of Technology Lausanne (EPFL)
Europe/Zurich timezone

From Foreground to Signal: Harnessing Deep Learning to Map Neutral Hydrogen During Reionization

Aug 29, 2024, 1:50 PM
20m
BC01 (level 0) (BC Building, Swiss Federal Institute of Technology Lausanne (EPFL))

BC01 (level 0)

BC Building, Swiss Federal Institute of Technology Lausanne (EPFL)

Rte Cantonale, 1015 Lausanne, Switzerland
Presentation Epoch of Reionization (EoR) EoR Science

Speaker

Michele Bianco (ETH Zurich)

Description

The next generation radio interferometry experiments will be sensible enough not only to detect the 21-cm signal but they will be able to map the distribution of neutral hydrogen during reionization and produce a tremendous amount of 3D tomographic data. The biggest challenge for the observational analysis of these images is to separate the 21-cm signal from the undesired foreground and instrumental noise contaminations.
Here, we present SERENEt (SEgmentation and REcovery NEtwork). A deep learning approach that works on SKA-Low mock observation with an observation time of 1000 h and in the presence of the Galactic synchrotron foreground. Our network identifies regions of neutral hydrogen (HI) and recovers the reionization 21-cm signal from those regions identified as neutral. We show that our approach can identify neutral regions during reionization with more than 87 percent accuracy and recover the 21-cm 2D power spectra with an average of 95 percent accuracy.

Primary author

Michele Bianco (ETH Zurich)

Presentation materials