arXiv subjects

Janis Fluri

Publications and source records attributed to Janis Fluri.

18 recordsLinked to original sources

arXiv2026

Emulating the complex galactic-scale orbital dynamics of LISA massive black hole pairs with normalizing flows

Massive black hole (MBH) pairs, formed in galaxy mergers, may coalesce in a burst of gravitational waves. Estimating the coalescence time-scales and rates is a long-standing astrophysical problem, essential to inform predictions for future gravitational-wave detectors such as the Laser Interferometer Space Antenna, but remains challenging: MBH orbital decay in realistic galactic environments is complex and stochastic, and non-axisymmetric structures such as stellar bars can perturb MBH pair dynamics, delaying or accelerating binary formation and undermining the assumption that dynamical friction alone sets the inspiral duration. Capturing this evolution requires simulations too expensive to run at population scale. Here we present an artificial-intelligence framework that emulates the galactic-scale orbital decay of an inspiralling MBH using conditional normalizing flows trained on a large suite of semi-analytical orbital integrations. Our model captures the evolution of secondary MBHs orbiting within multi-component galactic merger remnants featuring rotating stellar discs and bars, across a broad range of MBH masses, orbital configurations, and bar properties. The trained emulator reproduces the simulations' decay-time distributions while reducing computational cost by orders of magnitude. For the first time, we apply this model to galaxy populations drawn from a cosmological simulation, exploiting morphological information on barred and non-barred galaxies to infer MBH binary formation time-scales across cosmic time. Our results show that stellar bars can alter the distribution of MBH binary formation times. More broadly, this demonstrates how simulation-based, surrogate machine-learning emulators can unlock a class of astrophysical problems where the physics is well understood system-by-system but intractable at scale.

astro-ph.GA↗

arXiv2025

Scaling Behavior of Discrete Diffusion Language Models

Modern LLM pre-training consumes vast amounts of compute and training data, making the scaling behavior, or scaling laws, of different models a key distinguishing factor. Discrete diffusion language models (DLMs) have been proposed as an alternative to autoregressive language models (ALMs). However, their scaling behavior has not yet been fully explored, with prior work suggesting that they require more data and compute to match the performance of ALMs. We study the scaling behavior of DLMs on different noise types by smoothly interpolating between masked and uniform diffusion while paying close attention to crucial hyperparameters such as batch size and learning rate. Our experiments reveal that the scaling behavior of DLMs strongly depends on the noise type and is considerably different from ALMs. While all noise types converge to similar loss values in compute-bound scaling, we find that uniform diffusion requires more parameters and less data for compute-efficient training compared to masked diffusion, making them a promising candidate in data-bound settings. We scale our uniform diffusion model up to 10B parameters trained for $10^{22}$ FLOPs, confirming the predicted scaling behavior and making it the largest publicly known uniform diffusion model to date.

cs.LG↗

arXiv2025

Generalized Interpolating Discrete Diffusion

While state-of-the-art language models achieve impressive results through next-token prediction, they have inherent limitations such as the inability to revise already generated tokens. This has prompted exploration of alternative approaches such as discrete diffusion. However, masked diffusion, which has emerged as a popular choice due to its simplicity and effectiveness, reintroduces this inability to revise words. To overcome this, we generalize masked diffusion, deriving a new family of general interpolating discrete diffusion (GIDD) which offers greater flexibility in the design of the noising processes. Leveraging a novel diffusion ELBO, we achieve compute-matched state-of-the-art performance in diffusion language modeling. Exploiting GIDD's flexibility, we explore a hybrid approach combining masking and uniform noise, leading to improved sample quality and unlocking the ability for the model to correct its own mistakes, an area where autoregressive models notoriously have struggled. Code: https://github.com/dvruette/gidd/

cs.CL↗

arXiv2022

CosmoGridV1: a simulated $w$CDM theory prediction for map-level cosmological inference

We present CosmoGridV1: a large set of lightcone simulations for map-level cosmological inference with probes of large scale structure. It is designed for cosmological parameter measurement based on Stage-III photometric surveys with non-Gaussian statistics and machine learning. CosmoGridV1 spans the $w$CDM model by varying $\Omega_m$, $\sigma_8$, $w_0$, $H_0$, $n_s$, $\Omega_b$, and assumes three degenerate neutrinos with $\sum m_\nu$ = 0.06 eV. This space is covered by 2500 grid points on a Sobol sequence. At each grid point, we run 7 simulations with PkdGrav3 and store 69 particle maps at nside=2048 up to $z$=3.5, as well as halo catalog snapshots. The fiducial cosmology has 200 independent simulations, along with their stencil derivatives. An important part of CosmoGridV1 is the benchmark set of 28 simulations, which include larger boxes, higher particle counts, and higher redshift resolution of shells. They allow for testing if new types of analyses are sensitive to choices made in CosmoGridV1. We add baryon feedback effects on the map level, using shell-based baryon correction model. The shells are used to create maps of weak gravitational lensing, intrinsic alignment, and galaxy clustering, using the UFalcon code. The main part of CosmoGridV1 are the raw particle count shells that can be used to create full-sky maps for a given $n(z)$. We also release projected maps for a Stage-III forecast, as well as maps used previously in KiDS-1000 deep learning constraints with CosmoGridV1. The data is available at www.cosmogrid.ai.

astro-ph.CO↗

arXiv2022

Assessing theoretical uncertainties for cosmological constraints from weak lensing surveys

$ $Weak gravitational lensing is a powerful probe which is used to constrain the standard cosmological model and its extensions. With the enhanced statistical precision of current and upcoming surveys, high accuracy predictions for weak lensing statistics are needed to limit the impact of theoretical uncertainties on cosmological parameter constraints. For this purpose, we present a comparison of the theoretical predictions for the nonlinear matter and weak lensing power spectra, based on the widely used fitting functions ($\texttt{mead}$ and $\texttt{rev-halofit}$), emulators ($\texttt{EuclidEmulator}$, $\texttt{EuclidEmulator2}$, $\texttt{BaccoEmulator}$ and $\texttt{CosmicEmulator}$) and N-body simulations ($\texttt{Pkdgrav3}$). We consider the forecasted constraints on the $\Lambda \texttt{CDM}$ and $\texttt{wCDM}$ models from weak lensing for stage III and stage IV surveys. We study the relative bias on the constraints and their dependence on the assumed prescriptions. Assuming a $\Lambda \texttt{CDM}$ cosmology, we find that the relative agreement on the $S_8$ parameter is between $0.2-0.3\sigma$ for a stage III-like survey between the above predictors. For a stage IV-like survey the agreement becomes $1.4-3.0\sigma$. In the $\texttt{wCDM}$ scenario, we find broader $S_8$ constraints, and agreements of $0.18-0.26\sigma$ and $0.7-1.7\sigma$ for stage III and stage IV surveys, respectively. The accuracies of the above predictors therefore appear adequate for stage III surveys, while the fitting functions would need improvements for future stage IV weak lensing surveys. Furthermore, we find that, of the fitting functions, $\texttt{mead}$ provides the best agreement with the emulators. We discuss the implication of these findings for the preparation of the future weak lensing surveys.

astro-ph.CO↗

arXiv2022

Towards a full $w$CDM map-based analysis for weak lensing surveys

The next generation of weak lensing surveys will measure the matter distribution of the local Universe with unprecedented precision, allowing the resolution of non-Gaussian features of the convergence field. This encourages the use of higher-order mass-map statistics for cosmological parameter inference. We extend the forward-modelling based methodology introduced in a previous forecast paper to match these new requirements. We provide multiple forecasts for the wCDM parameter constraints that can be expected from stage 3 and 4 weak lensing surveys. We consider different survey setups, summary statistics and mass map filters including wavelets. We take into account the shear bias, photometric redshift uncertainties and intrinsic alignment. The impact of baryons is investigated and the necessary scale cuts are applied. We compare the angular power spectrum analysis to peak and minima counts as well as Minkowski functionals of the mass maps. We find a preference for Starlet over Gaussian filters. Our results suggest that using a survey setup with 10 instead of 5 tomographic redshift bins is beneficial. Adding cross-tomographic information improves the constraints on cosmology and especially on galaxy intrinsic alignment for all statistics. In terms of constraining power, we find the angular power spectrum and the peak counts to be equally matched for stage 4 surveys, followed by minima counts and the Minkowski functionals. Combining different summary statistics significantly improves the constraints and compensates the stringent scale cuts. We identify the most `cost-effective' combination to be the angular power spectrum, peak counts and Minkowski functionals following Starlet filtering.

astro-ph.CO↗

arXiv2022

DeepLSS: breaking parameter degeneracies in large scale structure with deep learning analysis of combined probes

In classical cosmological analysis of large scale structure surveys with 2-pt functions, the parameter measurement precision is limited by several key degeneracies within the cosmology and astrophysics sectors. For cosmic shear, clustering amplitude $\sigma_8$ and matter density $\Omega_m$ roughly follow the $S_8=\sigma_8(\Omega_m/0.3)^{0.5}$ relation. In turn, $S_8$ is highly correlated with the intrinsic galaxy alignment amplitude $A_{\rm{IA}}$. For galaxy clustering, the bias $b_g$ is degenerate with both $\sigma_8$ and $\Omega_m$, as well as the stochasticity $r_g$. Moreover, the redshift evolution of IA and bias can cause further parameter confusion. A tomographic 2-pt probe combination can partially lift these degeneracies. In this work we demonstrate that a deep learning analysis of combined probes of weak gravitational lensing and galaxy clustering, which we call DeepLSS, can effectively break these degeneracies and yield significantly more precise constraints on $\sigma_8$, $\Omega_m$, $A_{\rm{IA}}$, $b_g$, $r_g$, and IA redshift evolution parameter $\eta_{\rm{IA}}$. The most significant gains are in the IA sector: the precision of $A_{\rm{IA}}$ is increased by approximately 8x and is almost perfectly decorrelated from $S_8$. Galaxy bias $b_g$ is improved by 1.5x, stochasticity $r_g$ by 3x, and the redshift evolution $\eta_{\rm{IA}}$ and $\eta_b$ by 1.6x. Breaking these degeneracies leads to a significant gain in constraining power for $\sigma_8$ and $\Omega_m$, with the figure of merit improved by 15x. We give an intuitive explanation for the origin of this information gain using sensitivity maps. These results indicate that the fully numerical, map-based forward modeling approach to cosmological inference with machine learning may play an important role in upcoming LSS surveys. We discuss perspectives and challenges in its practical deployment for a full survey analysis.

astro-ph.CO↗

arXiv2022

A Full $w$CDM Analysis of KiDS-1000 Weak Lensing Maps using Deep Learning

We present a full forward-modeled $w$CDM analysis of the KiDS-1000 weak lensing maps using graph-convolutional neural networks (GCNN). Utilizing the $\texttt{CosmoGrid}$, a novel massive simulation suite spanning six different cosmological parameters, we generate almost one million tomographic mock surveys on the sphere. Due to the large data set size and survey area, we perform a spherical analysis while limiting our map resolution to $\texttt{HEALPix}$ $n_\mathrm{side}=512$. We marginalize over systematics such as photometric redshift errors, multiplicative calibration and additive shear bias. Furthermore, we use a map-level implementation of the non-linear intrinsic alignment model along with a novel treatment of baryonic feedback to incorporate additional astrophysical nuisance parameters. We also perform a spherical power spectrum analysis for comparison. The constraints of the cosmological parameters are generated using a likelihood free inference method called Gaussian Process Approximate Bayesian Computation (GPABC). Finally, we check that our pipeline is robust against choices of the simulation parameters. We find constraints on the degeneracy parameter of $S_8 \equiv \sigma_8\sqrt{\Omega_M/0.3} = 0.78^{+0.06}_{-0.06}$ for our power spectrum analysis and $S_8 = 0.79^{+0.05}_{-0.05}$ for our GCNN analysis, improving the former by 16%. This is consistent with earlier analyses of the 2-point function, albeit slightly higher. Baryonic corrections generally broaden the constraints on the degeneracy parameter by about 10%. These results offer great prospects for full machine learning based analyses of on-going and future weak lensing surveys.

astro-ph.CO↗

arXiv2021

A tomographic spherical mass map emulator of the KiDS-1000 survey using conditional generative adversarial networks

Large sets of matter density simulations are becoming increasingly important in large-scale structure cosmology. Matter power spectra emulators, such as the Euclid Emulator and CosmicEmu, are trained on simulations to correct the non-linear part of the power spectrum. Map-based analyses retrieve additional non-Gaussian information from the density field, whether through human-designed statistics such as peak counts, or via machine learning methods such as convolutional neural networks. The simulations required for these methods are very resource-intensive, both in terms of computing time and storage. Map-level density field emulators, based on deep generative models, have recently been proposed to address these challenges. In this work, we present a novel mass map emulator of the KiDS-1000 survey footprint, which generates noise-free spherical maps in a fraction of a second. It takes a set of cosmological parameters $(\Omega_M, \sigma_8)$ as input and produces a consistent set of 5 maps, corresponding to the KiDS-1000 tomographic redshift bins. To construct the emulator, we use a conditional generative adversarial network architecture and the spherical CNN $\texttt{DeepSphere}$, and train it on N-body-simulated mass maps. We compare its performance using an array of quantitative comparison metrics: angular power spectra $C_\ell$, pixel/peaks distributions, $C_\ell$ correlation matrices, and Structural Similarity Index. Overall, the average agreement on these summary statistics is $<10\%$ for the cosmologies at the centre of the simulation grid, and degrades slightly on grid edges. Finally, we perform a mock cosmological parameter estimation using the emulator and the original simulation set. We find good agreement in these constraints, for both likelihood and likelihood-free approaches. The emulator is available at https://tfhub.dev/cosmo-group-ethz/models/kids-cgan/1.

astro-ph.CO↗

arXiv2021

Symbolic Implementation of Extensions of the $\texttt{PyCosmo}$ Boltzmann Solver

$\texttt{PyCosmo}$ is a Python-based framework for the fast computation of cosmological model predictions. One of its core features is the symbolic representation of the Einstein-Boltzmann system of equations. Efficient $\texttt{C/C++}$ code is generated from the $\texttt{SymPy}$ symbolic expressions making use of the $\texttt{sympy2c}$ package. This enables easy extensions of the equation system for the implementation of new cosmological models. We illustrate this with three extensions of the $\texttt{PyCosmo}$ Boltzmann solver to include a dark energy component with a constant equation of state, massive neutrinos and a radiation streaming approximation. We describe the $\texttt{PyCosmo}$ framework, highlighting new features, and the symbolic implementation of the new models. We compare the $\texttt{PyCosmo}$ predictions for the $\Lambda$CDM model extensions with $\texttt{CLASS}$, both in terms of accuracy and computational speed. We find a good agreement, to better than 0.1% when using high-precision settings and a comparable computational speed. Links to the Python Package Index (PyPI) page of the code release and to the PyCosmo Hub, an online platform where the package is installed, are available at: https://cosmology.ethz.ch/research/software-lab/PyCosmo.html.

astro-ph.CO↗

arXiv2021

Combined $13\times2$-point analysis of the Cosmic Microwave Background and Large-Scale Structure: implications for the $S_8$-tension and neutrino mass constraints

We present cosmological constraints for the flat $\Lambda$CDM model, including the sum of neutrino masses, by performing a multi-probe analysis of a total of 13 tomographic auto- and cross-angular power spectra. This is achieved by combining, at map level, the latest primary CMB and CMB-lensing measurements from the Planck 2018 data release, as well as spectroscopic galaxy samples from BOSS DR12, and the latest Kilo-Degree Survey (KiDS-1000) tomographic weak lensing shear data release. Our analysis includes auto- and cross-correlations as well as calibration parameters for all cosmological probes, thus providing a self-calibration of the combined data sets. We find a good fit (reduced $\chi^2$=1.7) for the combined probes with calibration parameters only moderately different from their nominal value, thus giving a possible interpretation of the tension between the early- and late-Universe probes. The resulting value for the structure growth parameter is $S_8 = 0.754 \pm 0.016$ (68\% CL). We also obtain a $\sim$2.3$\sigma$ constraint on the neutrino mass sum of $\sum m_\nu = 0.51^{+0.21}_{-0.24}$ eV (68\% CL), which is compatible with current particle physics limits. We perform several tests by fixing the neutrino mass sum to a low value, considering narrower priors on the multiplicative bias parameters for cosmic shear, and by fixing all calibration parameters to their expected values. These tests result in worse fits compared to our fiducial run, especially for the case when all calibration parameters are fixed. This latter test also yields a lower upper limit of the neutrino mass sum. We discuss how the interplay between the cosmological and calibration parameters impact the $S_8$-tension and the constraints on the neutrino mass sum. [abridged]

astro-ph.CO↗

arXiv2021

Cosmological Parameter Estimation and Inference using Deep Summaries

The ability to obtain reliable point estimates of model parameters is of crucial importance in many fields of physics. This is often a difficult task given that the observed data can have a very high number of dimensions. In order to address this problem, we propose a novel approach to construct parameter estimators with a quantifiable bias using an order expansion of highly compressed deep summary statistics of the observed data. These summary statistics are learned automatically using an information maximising loss. Given an observation, we further show how one can use the constructed estimators to obtain approximate Bayes computation (ABC) posterior estimates and their corresponding uncertainties that can be used for parameter inference using Gaussian process regression even if the likelihood is not tractable. We validate our method with an application to the problem of cosmological parameter inference of weak lensing mass maps. We show in that case that the constructed estimators are unbiased and have an almost optimal variance, while the posterior distribution obtained with the Gaussian process regression is close to the true posterior and performs better or equally well than comparable methods.

astro-ph.CO↗

arXiv2020

Fast Lightcones for Combined Cosmological Probes

The combination of different cosmological probes offers stringent tests of the $\Lambda$CDM model and enhanced control of systematics. For this purpose, we present an extension of the lightcone generator UFalcon first introduced in Sgier et al. 2019 (arXiv:1801.05745), enabling the simulation of a self-consistent set of maps for different cosmological probes. Each realization is generated from the same underlying simulated density field, and contains full-sky maps of different probes, namely weak lensing shear, galaxy overdensity including RSD, CMB lensing, and CMB temperature anisotropies from the ISW effect. The lightcone generation performed by UFalcon is parallelized and based on the replication of a large periodic volume simulated with the GPU-accelerated $N$-Body code PkdGrav3. The post-processing to construct the lightcones requires only a runtime of about 1 walltime-hour corresponding to about 100 CPU-hours. We use a randomization procedure to increase the number of quasi-independent full-sky UFalcon map-realizations, which enables us to compute an accurate multi-probe covariance matrix. Using this framework, we forecast cosmological parameter constraints by performing a multi-probe likelihood analysis for a combination of simulated future stage-IV-like surveys. We find that the inclusion of the cross-correlations between the probes significantly increases the information gain in the parameter constraints. We also find that the use of a non-Gaussian covariance matrix is increasingly important, as more probes and cross-correlation power spectra are included. A version of the UFalcon package currently including weak gravitational lensing is publicly available.

astro-ph.CO↗

arXiv2020

Cosmological Forecast for non-Gaussian Statistics in large-scale weak Lensing Surveys

Cosmic shear data contains a large amount of cosmological information encapsulated in the non-Gaussian features of the weak lensing mass maps. This information can be extracted using non-Gaussian statistics. We compare the constraining power in the $\Omega_{\mathrm{m}} - \sigma_8$ plane of three map-based non-Gaussian statistics with the angular power spectrum, namely; peak/minimum counts and Minkowski functionals. We further analyze the impact of tomography and systematic effects originating from galaxy intrinsic alignments, multiplicative shear bias and photometric redshift systematics. We forecast the performance of the statistics for a stage-3-like weak lensing survey and restrict ourselves to scales $\geq$ 10 arcmin. We find, that in our setup, the considered non-Gaussian statistics provide tighter constraints than the angular power spectrum. The peak counts show the greatest potential, increasing the Figure-of-Merit (FoM) in the $\Omega_{\mathrm{m}} - \sigma_8$ plane by a factor of about 4. A combined analysis using all non-Gaussian statistics in addition to the power spectrum increases the FoM by a factor of 5 and reduces the error on $S_8$ by $\approx$ 25\%. We find that the importance of tomography is diminished when combining non-Gaussian statistics with the angular power spectrum. The non-Gaussian statistics indeed profit less from tomography and the minimum counts and Minkowski functionals add some robustness against galaxy intrinsic alignment in a non-tomographic setting. We further find that a combination of the angular power spectrum and the non-Gaussian statistics allows us to apply conservative scale cuts in the analysis, thus helping to minimize the impact of baryonic and relativistic effects, while conserving the cosmological constraining power. We make the code that was used to conduct this analysis publicly available.

astro-ph.CO↗

arXiv2019

Cosmological constraints with deep learning from KiDS-450 weak lensing maps

Convolutional Neural Networks (CNN) have recently been demonstrated on synthetic data to improve upon the precision of cosmological inference. In particular they have the potential to yield more precise cosmological constraints from weak lensing mass maps than the two-point functions. We present the cosmological results with a CNN from the KiDS-450 tomographic weak lensing dataset, constraining the total matter density $\Omega_m$, the fluctuation amplitude $\sigma_8$, and the intrinsic alignment amplitude $A_{\rm{IA}}$. We use a grid of N-body simulations to generate a training set of tomographic weak lensing maps. We test the robustness of the expected constraints to various effects, such as baryonic feedback, simulation accuracy, different value of $H_0$, or the lightcone projection technique. We train a set of ResNet-based CNNs with varying depths to analyze sets of tomographic KiDS mass maps divided into 20 flat regions, with applied Gaussian smoothing of $\sigma=2.34$ arcmin. The uncertainties on shear calibration and $n(z)$ error are marginalized in the likelihood pipeline. Following a blinding scheme, we derive constraints of $S_8 = \sigma_8 (\Omega_m/0.3)^{0.5} = 0.777^{+0.038}_{-0.036}$ with our CNN analysis, with $A_{\rm{IA}}=1.398^{+0.779}_{-0.724}$. We compare this result to the power spectrum analysis on the same maps and likelihood pipeline and find an improvement of about $30\%$ for the CNN. We discuss how our results offer excellent prospects for the use of deep learning in future cosmological data analysis.

astro-ph.CO↗

arXiv2018

Cosmological constraints from noisy convergence maps through deep learning

Deep learning is a powerful analysis technique that has recently been proposed as a method to constrain cosmological parameters from weak lensing mass maps. Due to its ability to learn relevant features from the data, it is able to extract more information from the mass maps than the commonly used power spectrum, and thus achieve better precision for cosmological parameter measurement. We explore the advantage of Convolutional Neural Networks (CNN) over the power spectrum for varying levels of shape noise and different smoothing scales applied to the maps. We compare the cosmological constraints from the two methods in the $\Omega_M-\sigma_8$ plane for sets of 400 deg$^2$ convergence maps. We find that, for a shape noise level corresponding to 8.53 galaxies/arcmin$^2$ and the smoothing scale of $\sigma_s = 2.34$ arcmin, the network is able to generate 45% tighter constraints. For smaller smoothing scale of $\sigma_s = 1.17$ the improvement can reach $\sim 50 \%$, while for larger smoothing scale of $\sigma_s = 5.85$, the improvement decreases to 19%. The advantage generally decreases when the noise level and smoothing scales increase. We present a new training strategy to train the neural network with noisy data, as well as considerations for practical applications of the deep learning approach.

astro-ph.CO↗

arXiv2018

Weak lensing peak statistics in the era of large scale cosmological surveys

Weak lensing peak counts are a powerful statistical tool for constraining cosmological parameters. So far, this method has been applied only to surveys with relatively small areas, up to several hundred square degrees. As future surveys will provide weak lensing datasets with size of thousands of square degrees, the demand on the theoretical prediction of the peak statistics will become heightened. In particular, large simulations of increased cosmological volume are required. In this work, we investigate the possibility of using simulations generated with the fast Comoving-Lagrangian acceleration (COLA) method, coupled to the convergence map generator Ufalcon, for predicting the peak counts. We examine the systematics introduced by the COLA method by comparing it with a full TreePM code. We find that for a 2000 deg$^2$ survey, the systematic error is much smaller than the statistical error. This suggests that the COLA method is able to generate promising theoretical predictions for weak lensing peaks. We also examine the constraining power of various configurations of data vectors, exploring the influence of splitting the sample into tomographic bins and combining different smoothing scales. We find the combination of smoothing scales to have the most constraining power, improving the constraints on the $S_8$ amplitude parameter by at least 40% compared to a single smoothing scale, with tomography brining only limited increase in measurement precision.

astro-ph.CO↗

arXiv2018

Fast cosmic web simulations with generative adversarial networks

Dark matter in the universe evolves through gravity to form a complex network of halos, filaments, sheets and voids, that is known as the cosmic web. Computational models of the underlying physical processes, such as classical N-body simulations, are extremely resource intensive, as they track the action of gravity in an expanding universe using billions of particles as tracers of the cosmic matter distribution. Therefore, upcoming cosmology experiments will face a computational bottleneck that may limit the exploitation of their full scientific potential. To address this challenge, we demonstrate the application of a machine learning technique called Generative Adversarial Networks (GAN) to learn models that can efficiently generate new, physically realistic realizations of the cosmic web. Our training set is a small, representative sample of 2D image snapshots from N-body simulations of size 500 and 100 Mpc. We show that the GAN-generated samples are qualitatively and quantitatively very similar to the originals. For the larger boxes of size 500 Mpc, it is very difficult to distinguish them visually. The agreement of the power spectrum $P_k$ is 1-2\% for most of the range, between $k=0.06$ and $k=0.4$. An important advantage of generating cosmic web realizations with a GAN is the considerable gains in terms of computation time. Each new sample generated by a GAN takes a fraction of a second, compared to the many hours needed by traditional N-body techniques. We anticipate that the use of generative models such as GANs will therefore play an important role in providing extremely fast and precise simulations of cosmic web in the era of large cosmological surveys, such as Euclid and Large Synoptic Survey Telescope (LSST).

astro-ph.CO↗