DeepMass

Cosmological map inference with deep learning

DeepMass was developed for inferring dark matter maps from weak gravitational lensing measurements, and uses deep learning to reconstruct cosmological maps.

DeepMass can also be incorporated into a Moment Network (see ArXiv:2011.05991) enabling high-dimensional likelihood-free inference.

(Dark matter mass map demo DES_mass_maps_demo/Training_example)

CMB result from ``Single frequency CMB B-mode inference with realistic foregrounds from a single training image'' Jeffrey et al. 2021 MNRAS Letters

(CMB Foreground demo CMB_foreground_demo/MomentNetwork_foregrounds)

Installation

To download data associated with the demos, this repository uses Git Large File Storage (git-lfs): https://git-lfs.github.com/

If this is not installed locally, the downloaded repository will include code but not data.

Using pip

!pip install 'git+https://github.com/NiallJeffrey/DeepMass.git'

From source

python setup.py install 

or for a cluster:

python setup.py install --user

Prerequisites

Python 3; Tensorflow>=2.2; healpy

Running the tests

python unit_tests.py

Authors

• Niall Jeffrey
• Francois Lanusse

License

This project is licensed under the MIT License - see the LICENSE.md file for details