Cosmology in the Machine Learning Era Francisco (Paco) Villaescusa-Navarro
Waterloo Center for Astrophysics October/21/2020 Outline
• Motivation
• Small scales: benefits and problems
• A machine learning approach
• Our vision Motivation: the LCDM model
{Ω , Ω , ℎ, � , � , � , � , � , � … } Goal: smallest error on the parameters
Ω ± �Ω Ω ± �Ω Answer fundamental questions: ℎ ± �ℎ
� ± �� 1) What is the nature of dark energy?
� ± �� 2) What are the neutrino masses? � ± �� 3) How fast is the Universe expanding? � ± �� 4) What are the properties of the Universe initial conditions?
� ± �� 5) … � ± �� Parameter inference
Observations Theory
�"(�) �!(�|�⃗)
What summary statistics shall we use to determine �⃗ with the smallest error? Parameter inference: summary statistics
Gaussian density field Fully described by the power spectrum
Non-Gaussian density field Mathematically “intractable” P(k), B(k), peaks, voids, … Summary • Goal: constrain cosmological parameters with the highest accuracy Parameter inference: information content
How well can we constraint some parameters �⃗ = {Ω , Ω , ℎ, � , � , � } given some observables �⃗ = {P k , P k , P k … P k , … }?
��⃗ ��⃗ F = � �� �� Fisher matrix