Résumé / Abstract Journal-club_Univers

Séminaire Univers /
Seminar Universe

« LyAl-Net: A high-efficiency Lyman-a forest simulation with neural network »

Chotipan Boonkongkird
Institut d'Astrophysique de Paris (Paris, France)

We present a machine learning approach to emulate the hydrodynamical simulation of intergalactic medium (IGM) physics for the Lyman-alpha forest called LyAl-Net. This technique could have a decisive impact on the results derived from present experiments, such as QSO surveys derived from SDSS3 and SDSS4 data. However, it could be critical for the exploitation of upcoming surveys like WEAVE-QSO, for which R=5000 and R=20000 in low and high-resolution modes, respectively. This work uses a neural network based on the U-net architecture to predict neutral hydrogen physical properties, density, and temperature. We have used only 9\% of the volume covered by the Horizon-noAGN simulation to train the LyAl-Net. More generally, the computation of individual fields from the dark matter density agrees well with regular physical regimes of cosmological fields. The mean transmission calculated from the predicted quantities performed is 2.5% error to the one derived from the raw fields of the original simulation. We also explored the transfer learning framework for other cosmological simulations containing different Baryonic feedbacks. We test the concept by correcting the equation of the state of IGM. The results tested on IllustrisTNG100 showed a drastic improvement.
mardi 25 octobre 2022 - 11:00
Salle des séminaires Évry Schatzman
Institut d'Astrophysique de Paris
Pages web du séminaire / Seminar's webpage