Speaker
Description
The next generation of galaxy surveys will provide unprecedented data, leading to accurate tests of gravity on cosmological scales. To fully exploit the nonlinear information encoded in the large-scale structure of the Universe, we propose to leverage cutting-edge deep learning algorithms, such as diffusion models, to efficiently generate 3D density fields conditioned on cosmological parameters. This approach is capable of fast and accurate emulation of cosmic volumes, while maintaining consistency with summary statistics and achieving a low computational cost comparable to state-of-the-art N-body simulations of modified gravity cosmologies. We demonstrate that trained diffusion models can be used to derive robust and accurate constraints on cosmological parameters, offering an efficient alternative for cosmological analysis with the same accuracy as traditional methods.
