A GPU-Accelerated JAX Framework for Robust Parametric Component Separation and Clustering Optimization for CMB Polarization Satellites
This work introduces a JAX-powered framework to separate CMB polarization data by accounting for spatially varying foregrounds. It uses thousands of pixel configurations, optimized via K-means clustering, to find the best balance between model complexity and noise. This method significantly improves the accuracy of estimating cosmological parameters like the tensor-to-scalar ratio.