Tools WG session of GDR Cophy

Nov 18, 2024, 1:55 PM → Nov 20, 2024, 12:00 PM Europe/Paris

IAP

Guilhem Lavaux (IAP), Mickael Rigault (IPNL), Simon Prunet (OCA), Yann Rasera (LUTH/Université Paris Cité/Observatoire de Paris/IUF)

This working group (website here) is dedicated to raise awareness of new tools and methodology developed in the wider French community to do cosmology. The tools must be understood in a wide sense: it may be large software stack, online websites, or Mathematica© notebook.

This Working Group aims at sharing and helping the development of tools for which the needs are common to all the actors of the GDR.

It includes:

• tools that are oriented toward the predictions of cosmological observables (including non-linearities) and their relative theoretical systematic uncertainties such as Boltzmann solvers, numerical simulations, approximate methods for the non-linear evolution, and forward modeling,
• development of estimators for data analysis which are for example based on power spectra, and higher order statistics;
• mathematical methods for inference such as the one based on MCMC, Machine Learning, and minimizer;
• production of validation data, such as mock galaxy catalog, mock CMB maps;
• tools for symbolic-based computation in the cosmological context.

This session is dedicated particularly to the use of JAX, PyTorch, and others in cosmology.

The meeting will happen in salle des séminaires / Evry Schatzman (level -1) at the IAP on November 18-20 2024.

Please note that the present timetable is preliminary and slots may be swapped depending on external cosntraints.

 

Monday, November 18

1:55 PM → 2:00 PM Organisation

1:55 PM Welcome

Speaker: Guilhem Lavaux (IAP)

2:00 PM → 3:30 PM New libraries

2:00 PM JAX basics

Speaker: Jean-Eric Campagne (IJCLab)

intro_JAX_for_cophy.pdf

JAX Tuto for GDR Cophy Tools

3:30 PM → 4:00 PM Coffee break

4:00 PM → 5:30 PM New libraries

4:00 PM Massively Parallel Computing in Cosmology with JAX

Speaker: Wassim Kabalan (APC)

Slides

Solutions

Tutorials

7:30 PM → 9:30 PM Diner

Tuesday, November 19

9:30 AM → 10:30 AM New possibilities

9:30 AM CMB likelihood

Speaker: Lennart Balkenhol (IAP)

candl_cophy_tools_compressed.pdf

10:30 AM → 11:00 AM Coffee break

11:00 AM → 12:30 PM New tools

11:00 AM LTU-Implicit Likelihood Inference

Speaker: Deaglan Bartlett (Institut d'Astrophysique de Paris)

Example 1 - Toy Problem

Example 2 - CAMELS

Example 3 - Molecules with Graphs

Example 4 - Dust Inference

Example 5 - Cluster Mass Estimates

LtU-ILI Presentation

2:00 PM → 3:30 PM New libraries

2:00 PM Probabilistic Programming for Cosmology with JAX

Speaker: Hugo Simon Onfroy (CEA Paris-Saclay)

PPL4Cosmo

slides

3:30 PM → 4:00 PM Coffee break

4:00 PM → 5:30 PM New possibilities

4:00 PM Lensing modeling with JAX in LSST

Speaker: Justine Zeghal (APC, CNRS)

4:45 PM Fast likelihood for SN Ia using JAX

Speaker: Marc Betoule (LPNHE)

Wednesday, November 20

9:50 AM → 10:10 AM New possibilities

9:50 AM JAX-Healpy: a JAX package compatible with the Healpy library

The Healpy package is widely used in cosmology, both for the handling of the HEALPix pixelization and the associated spherical harmonic transforms. The current Healpy package implementation is not JAX friendly, as the spherical harmonics implementation is based in C++, which prevents us to possibly accelerate the associated routines using JIT or to run them on GPUs. In this talk, we will present JAX-Healpy, a package aimed at proposing Healpy routines implemented using JAX. In particular, we will review the current status of the codes, with the spherical harmonics implementation wrapping the JAX package S2FFT.

Speaker: Magdy Morshed (INFN Ferrara)

Slides

10:10 AM → 10:40 AM Coffee break

10:40 AM → 11:40 AM New possibilities

10:40 AM s2scat : a software in JAX for generative models on the sphere using Scattering Transforms

Scattering Transforms are a new type of summary statistics recently developed for the study of highly non-Gaussian processes. These statistics can be used to build generative data models, and even to develop new component separation techniques. In the context of upcoming cosmological surveys, such as LiteBIRD for the cosmic microwave background polarisation or the Vera C. Rubin Observatory and the Euclid space telescope for study of the large-scale structures of the Universe, extending these tools to spherical data is necessary. In this work, we developed s2scat, a software in JAX, for scattering transforms on the sphere and focused on the construction of maximum-entropy generative models of several astrophysical fields.

Speaker: Louise Mousset (LPENS)

GDR_Cophy_Nov2024.pdf

11:10 AM FURAX: a modular JAX toolbox for solving inverse problems in science

Modern scientific data analyses involve complex models, presenting significant challenges in both data volume and computation. We present FURAX (Framework for Unified and Robust data Analysis with JAX), an open-source Python library that provides building blocks to construct instrument and noise models in a modular fashion, that benefit from the cutting-edge optimisation and GPU utilisation from JAX. FURAX is applied to cosmological data analysis with the CMB data. The examples include maximum-likelihood map-making, gap-filling of a time-ordered series and incorporation of non-ideal instrumental components.

Speaker: Pierre Chanial (APC)

Slides