Deep Learning for Observational Cosmology – DLOC

Co-PIs:

Dr. Tomasz Kacprzak (ETH Zürich)

Dr. Aurelien Lucchi (ETH Zürich)

Prof. Alexandre Refregier (ETH Zürich)

Prof. Thomas Hofmann (ETH Zürich)

Problem:

Building on the latest developments in machine learning in order to address some major problems in cosmological analysis: 1. Observations are financially expensive 2. Simulations are computationally expensive

Solution:

  1. Improve the classifier
  • Use convolutional neural networks
  1. Replace a part of the simulation with AI
  • Use generative adversarial networks

Impact:

Cosmology science is heavily dependent on simulations and forward models. Algorithms, which can simulate cosmic structures had a large impact in and drove many branches of cosmology.