florentleclercq - Network

Mon séminaire au @collegedefrance est en ligne ! "L’énergie noire avec Euclid",...

Mon séminaire au @collegedefrance est en ligne ! "L’énergie noire avec Euclid", https://www.youtube.com/watch?v=bLy8_iKY0rE
Merci à Françoise Combes pour son invitation.
#astrophysics #darkenergy #euclid #cosmology

Le 13 janvier prochain, je serai au Collège de France @collegedefrance pour parler de l'une des grandes questions actuelles de...

Le 13 janvier prochain, je serai au Collège de France @collegedefrance pour parler de l'une des grandes questions actuelles de l'astrophysique : l'énergie noire. 🌌

Mon séminaire aura lieu lors de la dernière séance du cycle "Le secteur sombre de l’Univers : matière et énergie noires" de Françoise Combes, titulaire de la chaire Galaxies et Cosmologie. J'explorerai les différentes théories possibles et leurs implications pour l’avenir de l’Univers, en mettant l'accent sur les résultats que nous allons pouvoir obtenir grâce à Euclid @ec_euclid, la mission spatiale de l'ESA. 🛰️

📅 Quand ? Le 13 janvier 2025 à 17h45 📍 Où ? Amphithéâtre Marguerite de Navarre, Site Marcelin Berthelot, Collège de France, Paris (ou en ligne plus tard !)

#astrophysics #darkenergy #euclid #cosmology

3/3 How it works:1️⃣ Infer the initial power spectrum using SELFI,2️⃣ Use it to expose and diagnose misspecifications (galaxy bias,...

3/3 How it works:
1️⃣ Infer the initial power spectrum using SELFI,
2️⃣ Use it to expose and diagnose misspecifications (galaxy bias, masks, redshift errors, etc.) before parameter inference,
3️⃣ Recycle simulations for efficient cosmological parameter estimation.

Tristan Hoellinger & Florent Leclercq, Diagnosing systematic effects using the inferred initial power spectrum, https://arxiv.org/abs/2412.04443.
Developed in @AquilaScience at @astroIAP.

2/3 The next generation of galaxy surveys (DESI, Euclid @ec_euclid, LSST @VRubinObs) will revolutionise #cosmology—but only if we tackle...

2/3 The next generation of galaxy surveys (DESI, Euclid @ec_euclid, LSST @VRubinObs) will revolutionise #cosmology—but only if we tackle *systematic uncertainties*. Our new method builds on SELFI (https://arxiv.org/abs/1902.10149) & Implicit Likelihood Inference of Bayesian hierarchical models (https://arxiv.org/abs/2209.11057) to detect and avoid hidden model biases.

1/3 Diagnosing systematic effects in galaxy surveys before inferring cosmology just got easier (https://arxiv.org/abs/2412.04443). Using the...

1/3 Diagnosing systematic effects in galaxy surveys before inferring cosmology just got easier (https://arxiv.org/abs/2412.04443). Using the inferred initial power spectrum, we uncover and correct model misspecifications that could bias (Ωm, σ8) by over 2σ. Our 2-step approach makes it possible to spot model errors before they mislead you. #cosmology #datascience

5/5 Tested on particle-mesh cosmological simulations, COCA reduces errors and requires fewer force evaluations than COLA, providing accurate...

5/5 Tested on particle-mesh cosmological simulations, COCA reduces errors and requires fewer force evaluations than COLA, providing accurate density and velocity fields. COCA outperforms direct emulation in accuracy and robustness, even out of the range of training data.
Deaglan J. Bartlett, Marco Chiarenza, Ludvig Doeser & Florent Leclercq, COmoving Computer Acceleration (COCA): N-body simulations in an emulated frame of reference. Developed in @AquilaScience at @astroIAP.

4/5 Calculating perturbations around the machine-learned solution makes simulations cheaper by skipping unnecessary force evaluations, while...

4/5 Calculating perturbations around the machine-learned solution makes simulations cheaper by skipping unnecessary force evaluations, while maintaining accuracy and asymptotic convergence. We solve the true equations of motion so are guaranteed to converge to the truth!

3/5 COCA involves neural networks to emulate a frame of reference (rather than the simulation output) in a COLA-like framework for N-body...

3/5 COCA involves neural networks to emulate a frame of reference (rather than the simulation output) in a COLA-like framework for N-body simulations of dark matter particles. As COCA solves physical equations in an emulated frame of reference, it corrects emulation errors by design.

2/5 Interpretability and accuracy are pivotal challenges in the application of machine learning to cosmology. If machines find something...

2/5 Interpretability and accuracy are pivotal challenges in the application of machine learning to cosmology. If machines find something humans don't understand, how can we check (and trust) the results? In this paper, we contend that addressing this concern is not always obligatory, when machine learning is used to emulate gravitational N-body simulations.

1/5 At the last COSMO21 conference in Chania, I gave a talk titled "Don't trust neural networks? Me neither, but here's how I...

1/5 At the last COSMO21 conference in Chania, I gave a talk titled "Don't trust neural networks? Me neither, but here's how I use them anyway" (https://florent-leclercq.eu/talks/2024-05-21_COSMO21_Chania.pdf). The paper is now out on arXiv: https://arxiv.org/abs/2409.02154. #MachineLearning #Cosmology

Ma conférence de l'été dernier au 33ème Festival d'#Astronomie de Fleurance @AstroFleurance est en ligne : "La satellite...

Ma conférence de l'été dernier au 33ème Festival d'#Astronomie de Fleurance @AstroFleurance est en ligne : "La satellite @euclid_fr et l'analyse de ses données : énergie noire, information et incertitude"
https://www.youtube.com/watch?v=5Nt7CT0ozxg @ec_euclid #cosmologie #IntelligenceArtificielle