Baryonic processes that alter the large-scale distribution of gas, and thus the matter power spectrum, such as AGN feedback, are one of the the main systematics in current and future weak lensing surveys. Left uncorrected, these effects will bias the inferred properties of dark matter and dark energy that these surveys are designed to measure. Characterising the distribution of gas is thus of vital importance if these surveys are to be exploited to their full potential.
In this talk, I will present ongoing work on joint analyses of weak lensing and tracers of diffuse gas, specifically the tSZ effect. I will show how a joint analysis of cosmic shear and tSZ cross-correlations breaks degeneracies of the individual probes and can significantly improve the constraints on cosmological parameters compared to cosmic shear alone.
Finally, I will demonstrate how we use a class of machine learning methods - deep generative models - to augment N-body simulations with gas. Specifically, I will show how conditional variational autoencoders and generative adversarial networks trained on the BAHAMAS hydrodynamical simulations can be used "paint" pressure fields on the SLICS suite of N-body simulations to produce consistent lensing and tSZ maps for use in the estimation of the cross-correlation covariance.
Add to Calendar
2019-09-30T15:00:002019-09-30T16:00:00Weak lensing, baryons, and deep learningEvent Information:
Baryonic processes that alter the large-scale distribution of gas, and thus the matter power spectrum, such as AGN feedback, are one of the the main systematics in current and future weak lensing surveys. Left uncorrected, these effects will bias the inferred properties of dark matter and dark energy that these surveys are designed to measure. Characterising the distribution of gas is thus of vital importance if these surveys are to be exploited to their full potential.
In this talk, I will present ongoing work on joint analyses of weak lensing and tracers of diffuse gas, specifically the tSZ effect. I will show how a joint analysis of cosmic shear and tSZ cross-correlations breaks degeneracies of the individual probes and can significantly improve the constraints on cosmological parameters compared to cosmic shear alone.
Finally, I will demonstrate how we use a class of machine learning methods - deep generative models - to augment N-body simulations with gas. Specifically, I will show how conditional variational autoencoders and generative adversarial networks trained on the BAHAMAS hydrodynamical simulations can be used "paint" pressure fields on the SLICS suite of N-body simulations to produce consistent lensing and tSZ maps for use in the estimation of the cross-correlation covariance.Event Location:
Hennings 318