The influence of systematic on galaxy catalogs - using graph neural networks to infer Omega_m - VIRTUAL

We recently showed that a powerful way to constrain cosmological parameters from galaxy catalogs (only using their positions and radial velocities) is to train graph neural networks (GNNs) to perform field-level likelihood-free inference without imposing cuts on scale. Nevertheless, various factors affect observations, including: 1) masking, 2) uncertainties in peculiar velocities and radial distances, 3) different galaxy selections. In this talk, I will present models trained and tested on galaxy catalogs generated from thousands of state-of-the-art hydrodynamic simulations, conducted with different codes as part of the CAMELS project. These catalogs incorporate the mentioned observational effects. The results indicate that while the presence of these effects reduces the precision and accuracy of the models and increases the instances where the model fails, it still performs well in over 90% of galaxy catalogs. This demonstrates the potential of these models to constrain cosmological parameters, even when applied to real data.

---

Zoom link TBA