<tt>matryoshka</tt> II: accelerating effective field theory analyses of the galaxy power spectrum

Donald-McCann, Jamie; Koyama, K.; Beutler, Florian

doi:10.1093/mnras/stac3326

Cited by 8 publications

(3 citation statements)

References 27 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…This is ideal for sampling posterior probabilities in Bayesian parameter estimation. Our method can be straightforwardly generalised to the multipoles of the spectra, and could also be combined with emulators that make predictions for the true clustering signal (including galaxy biasing) based on perturbation theory (e.g., Donald-McCann et al 2023;DeRose et al 2022;Eggemeier et al 2022). Additional corrections due to binning the theory predictions in exactly the same way as done for the measurements (see e.g., Sect.…”

Section: Discussionmentioning

confidence: 99%

Window function convolution with deep neural network models

Alkhanishvili

Porciani

Sefusatti

2023

A&A

View full text Add to dashboard Cite

Traditional estimators of the galaxy power spectrum and bispectrum are sensitive to the survey geometry. They yield spectra that differ from the true underlying signal since they are convolved with the window function of the survey. For the current and future generations of experiments, this bias is statistically significant on large scales. It is thus imperative that the effect of the window function on the summary statistics of the galaxy distribution is accurately modelled. Moreover, this operation must be computationally efficient in order to allow sampling posterior probabilities while performing Bayesian estimation of the cosmological parameters. In order to satisfy these requirements, we built a deep neural network model that emulates the convolution with the window function, and we show that it provides fast and accurate predictions. We trained (tested) the network using a suite of 2000 (200) cosmological models within the cold dark matter scenario, and demonstrate that its performance is agnostic to the precise values of the cosmological parameters. In all cases, the deep neural network provides models for the power spectra and the bispectrum that are accurate to better than 0.1% on a timescale of 10 μs.

show abstract

Section: Discussionmentioning

confidence: 99%

Window function convolution with deep neural network models

Alkhanishvili

Porciani

Sefusatti

2023

A&A

View full text Add to dashboard Cite

show abstract

“…X lie = X and y lie = y [11] repeat M times [12] find x add = argmax[a(x)] starting from n r,acq starting locations [13] X lie append x add and X new append x add [14] y lie append µ(x add ) Kriging believer [15] GP_fit(X lie , y lie ) [16] end [17] y true = log L(X new ) + log π(X new ) parallelizable [18] X append X new [19] y append y true [20] if is_converged (e.g. equation (4.4)) then break [21] end [22] Sample µ(x) with MC sampler [23] return MC sample [24] Function GP_fit(X, y) [25] Compute K −1 = k(X, X|θ MAP ) −1 matrix inversion [26] µ(x) = µ GP+SVM (x) equations (2.5) and (3.4) [27] σ(x) = Σ GP+SVM (x) equations (2.6) and (3.5)…”

Section: Jcap10(2023)021mentioning

confidence: 99%

Fast and robust Bayesian inference using Gaussian processes with GPry

El Gammal,

Schöneberg,

Torrado

et al. 2023

J. Cosmol. Astropart. Phys.

View full text Add to dashboard Cite

We present the GPry algorithm for fast Bayesian inference of general (non-Gaussian) posteriors with a moderate number of parameters. GPry does not need any pre-training, special hardware such as GPUs, and is intended as a drop-in replacement for traditional Monte Carlo methods for Bayesian inference. Our algorithm is based on generating a Gaussian Process surrogate model of the log-posterior, aided by a Support Vector Machine classifier that excludes extreme or non-finite values. An active learning scheme allows us to reduce the number of required posterior evaluations by two orders of magnitude compared to traditional Monte Carlo inference. Our algorithm allows for parallel evaluations of the posterior at optimal locations, further reducing wall-clock times. We significantly improve performance using properties of the posterior in our active learning scheme and for the definition of the GP prior. In particular we account for the expected dynamical range of the posterior in different dimensionalities. We test our model against a number of synthetic and cosmological examples. GPry outperforms traditional Monte Carlo methods when the evaluation time of the likelihood (or the calculation of theoretical observables) is of the order of seconds; for evaluation times of over a minute it can perform inference in days that would take months using traditional methods. GPry is distributed as an open source Python package (pip install gpry) and can also be found at https://github.com/jonaselgammal/GPry.

show abstract

“…On an average single-core CPU, the model can produce ∼ 5000 predictions in a second. The prediction speed of emulators is key to enabling fast cosmological inference with Monte Carlo techniques which require 10 5 -10 6 evaluations [66][67][68].…”

Section: Emulatormentioning

confidence: 99%

Fast production of cosmological emulators in modified gravity: the matter power spectrum

Fiorini,

Koyama,

Baker

2023

J. Cosmol. Astropart. Phys.

Self Cite

View full text Add to dashboard Cite

We test the convergence of fast simulations based on the COmoving Lagrangian Acceleration (COLA) method for predictions of the matter power spectrum, specialising our analysis in the redshift range 1 ≤ z ≤ 1.65, relevant to high-redshift spectroscopic galaxy surveys. We then focus on the enhancement of the matter power spectrum in modified gravity (MG), the boost factor, using the Dvali-Gabadadze-Porrati (DGP) theory as a test case but developing a general approach that can be applied to other MG theories. After identifying the minimal simulation requirements for accurate DGP boost factors, we design and produce a COLA simulation suite that we use to train a neural network emulator for the DGP boost factor. Using MG-AREPO simulations as a reference, we estimate the emulator accuracy to be of ∼ 3% up to k = 5 h Mpc-1 at 0 ≤ z ≤ 2. We make the emulator publicly available at: https://github.com/BartolomeoF/nDGPemu.

show abstract

`matryoshka` II: accelerating effective field theory analyses of the galaxy power spectrum

Cited by 8 publications

References 27 publications

Window function convolution with deep neural network models

Window function convolution with deep neural network models

Fast and robust Bayesian inference using Gaussian processes with GPry

Fast production of cosmological emulators in modified gravity: the matter power spectrum

Contact Info

Product

Resources

About

matryoshka II: accelerating effective field theory analyses of the galaxy power spectrum

Cited by 8 publications

References 27 publications

Window function convolution with deep neural network models

Window function convolution with deep neural network models

Fast and robust Bayesian inference using Gaussian processes with GPry

Fast production of cosmological emulators in modified gravity: the matter power spectrum

Contact Info

Product

Resources

About

`matryoshka` II: accelerating effective field theory analyses of the galaxy power spectrum