Orateur
Description
Accurate statistical characterization of Galactic foregrounds is crucial for extracting cosmological information from CMB polarization data. However, the complexity of the interstellar medium leads to non-Gaussian structures that cannot be fully captured by traditional summary statistics, like the power spectrum. On the other hand, simulation and machine-learning approaches, while informative, face limitations: the former rely on simplifying physical assumptions, and the latter require large volumes of realistic training data that are in general not available. Further complicating the problem, polarized foregrounds are never observed in isolation but are always mixed with nuisance signals, including the CMB and instrumental noise.
A promising alternative to traditional methods for the statistical characterization and modeling of the Galactic polarized foregrounds is provided by scattering transforms (ST). These are a mathematically grounded set of summary statistics that efficiently capture multiscale, non-Gaussian features of complex physical fields and can be robustly estimated from limited data.
In this talk, we will discuss recent progress in applying ST to Planck data in order to separate the Galactic polarized foregrounds from the CMB and instrumental noise. Crucially, we will show some first steps in using information across multiple frequency channels, and we will present maps of Galactic polarized foregrounds separated from CMB and instrumental noise. These results highlight the potential of ST in exploiting both spatial and spectral information in the task of Galactic polarized foreground modeling, laying the groundwork for a more complete generative modeling approach, that could subsequently be applied to more recent data, such as ACT, SPT, or SO.
