arXiv subjects

Kunhao Zhong

Publications and source records attributed to Kunhao Zhong.

11 recordsLinked to original sources

arXiv2025

Vision Transformers for Cosmological Fields: Application to Weak Lensing Mass Maps

Weak gravitational lensing is a powerful probe of the universe's growth history. While traditional two-point statistics capture only the Gaussian features of the convergence field, deep learning methods such as convolutional neural networks (CNNs) have shown promise in extracting non-Gaussian information from small-scale, nonlinear structures. In this work, we evaluate the effectiveness of attention-based architectures, including variants of vision transformers (ViTs) and shifted window (Swin) transformers, in constraining the cosmological parameters $Ω_m$ and $S_8$ from weak lensing mass maps. Using a simulation-based inference (SBI) framework, we compare transformer-based methods to CNNs. We also examine performance scaling with the number of available $N$-body simulations, highlighting the importance of pre-training for transformer architectures. We find that the Swin transformer performs significantly better than vanilla ViTs, especially with limited training data. Despite their higher representational capacity, the Figure of Merit for cosmology achieved by transformers is comparable to that of CNNs under realistic noise conditions.

astro-ph.CO↗

arXiv2025

Separating the Inseparable: Constraining Arbitrary Primordial Bispectra with Cosmic Microwave Background Data

To efficiently probe primordial non-Gaussianity using Cosmic Microwave Background (CMB) data, we require theoretical predictions that are factorizable, \textit{i.e.}\ those whose kinematic dependence can be separated. This property does not hold for many models, hindering their application to data. In this work, we introduce a general framework for constructing separable approximations to primordial bispectra, enabling direct CMB constraints on arbitrary models including those computed using numerical tools. In contrast to other approaches such as modal decompositions, we learn the basis functions directly from the data, allowing high-fidelity representations with just a handful of terms. This is practically implemented using machine-learning techniques, utilizing neural network basis functions and a loss function designed to mimic the CMB cosine similarity. We validate our pipeline using a variety of input bispectra, demonstrating that the approximations are $>99.5\%$ correlated with the truth with just three terms. By incorporating the neural basis into the \textsc{PolySpec} code, we derive KSW-type CMB estimators, which reproduce local- and equilateral-type non-Gaussianity to within $0.1σ$. As a proof-of-concept, we constrain two inflationary bispectra from the `cosmological collider' scenario; these feature an additional strongly-mixed particle sector and cannot be computed analytically. By combining the numerical predictions from \textsc{CosmoFlow} with our factorizable approach (with just three terms), we place novel constraints on the collider models using \textit{Planck} PR4 data, finding no detection of non-Gaussianity. Our method facilitates detailed studies of the inflationary paradigm, connecting modern theoretical tools with high-resolution observational data.

astro-ph.CO↗

arXiv2025

Tests of Evolving Dark Energy with Geometric Probes of the Late-Time Universe

Recent results from the Dark Energy Spectroscopic Instrument (DESI) have shown a strong statistical preference for a time-evolving dark energy model over $Λ$CDM when combining BAO, CMB, and supernova (SN) data. We investigate the robustness of this conclusion by isolating geometric information in weak lensing measurements from the DES Year 3 survey and combining it with different datasets. We introduce a hyperparameter, $Ω_{\rm m}^{\rm growth}$, to decouple the growth contribution from the lensing 2-point correlation and thus bypass the possible effect of the $σ_8$ tension in our analysis. We then combine with the late-time geometric probes provided by BAO and SN, along with CMB primary data. The preference for evolving dark energy is consistent with the DESI-DR2 findings: when combining BAO, primary CMB, and weak lensing data, the $w_0w_a$CDM is preferred at about the $3σ$ significance. However, when we add SN, the result is sensitive to the choice of data: if we leave out $z<0.1$ SN data in the analysis, as a test of the effect of inhomogeneous calibration, we obtain a statistical significance below $2σ$ for time evolving dark energy. Indeed, the high-z only SN data \textbf{lowers} the evidence for evolving dark energy in all the data combinations we have examined. This underscores the importance of improved SN samples at low redshift and of alternative data combinations. We show that cosmic shear measurements with LSST Year 1 data will provide comparable power to current SN data. We discuss other low-redshift probes provided by lensing and galaxy clustering to test for evolving dark energy.

astro-ph.CO↗

arXiv2025

Investigating Late-Time Dark Energy and Massive Neutrinos in Light of DESI Y1 BAO

Baryonic Acoustic Oscillation (BAO) data from the Dark Energy Spectroscopic Instrument (DESI), in combination with Cosmic Microwave Background (CMB) data and Type Ia Supernovae (SN) luminosity distances, suggests a dynamical evolution of the dark energy equation of state with a phantom phase ($w < -1$) in the past when the so-called $w_0w_a$ parametrization $w(a) = w_0 + w_a(1-a)$ is assumed. In this work, we investigate more general dark energy models that also allow a phantom equation of state. We consider three cases: an equation of state with a transition feature, a model-agnostic equation of state with constant values in chosen redshift bins, and a k-essence model. Since the dark energy equation of state is correlated with neutrino masses, we reassess constraints on the neutrino mass sum focusing on the model-agnostic equation of state. We find that the combination of DESI BAO with Planck 2018 CMB data and SN data from Pantheon, Pantheon+, or Union3 is consistent with an oscillatory dark energy equation of state, while a monotonic behavior is preferred by the DESY5 SN data. Performing model comparison techniques, we find that the $w_0w_a$ parametrization remains the simplest dark energy model that can provide a better fit to DESI BAO, CMB, and all SN datasets than $Λ$CDM. Constraints on the neutrino mass sum assuming dynamical dark energy are relaxed compared to $Λ$CDM and we show that these constraints are tighter in the model-agnostic case relative to $w_0w_a$ model by $70\%-90\%$.

astro-ph.CO↗

arXiv2024

Fast Generation of Weak Lensing Maps with Analytical Point Transformation Functions

Nonlinear cosmological fields like galaxy density and lensing convergence can be approximately related to Gaussian fields via analytic point transforms. The lognormal transform (LN) has been widely used and is a simple example of a function that relates nonlinear fields to Gaussian fields. We consider more accurate General Point-Transformed Gaussian (GPTG) functions for such a mapping and apply them to convergence maps. We show that we can create maps that preserve the LN's ability to exactly match any desired power spectrum but go beyond LN by significantly improving the accuracy of the probability distribution function (PDF). With the aid of symbolic regression, we find a remarkably accurate GPTG function for convergence maps: its higher-order moments, scattering wavelet transform, Minkowski functionals, and peak counts match those of N-body simulations to the statistical uncertainty expected from tomographic lensing maps of the Rubin LSST 10 years survey. Our five-parameter function performs 2 to 5$\times$ better than the lognormal. We restrict our study to scales above about 7 arcmin; baryonic feedback alters the mass distribution on smaller scales. We demonstrate that the GPTG can robustly emulate variations in cosmological parameters due to the simplicity of the analytic transform. This opens up several possible applications, such as field-level inference, rapid covariance estimation, and other uses based on the generation of arbitrarily many maps with laptop-level computation capability.

astro-ph.CO↗

arXiv2024

Attention-Based Neural Network Emulators for Multi-Probe Data Vectors Part II: Assessing Tension Metrics

The next generation of cosmological surveys is expected to generate unprecedented high-quality data, consequently increasing the already substantial computational costs of Bayesian statistical methods. This will pose a significant challenge to analyzing theoretical models of cosmology. Additionally, new mitigation techniques of baryonic effects, intrinsic alignment, and other systematic effects will inevitably introduce more parameters, slowing down the convergence of Bayesian analyses. In this scenario, machine-learning-based accelerators are a promising solution, capable of reducing the computational costs and execution time of such tools by order of thousands. Yet, they have not been able to provide accurate predictions over the wide prior ranges in parameter space adopted by Stage III/IV collaborations in studies employing real-space two-point correlation functions. This paper offers a leap in this direction by carefully investigating the modern transformer-based neural network (NN) architectures in realistic simulated Rubin Observatory year one cosmic shear $Λ$CDM inferences. Building on the framework introduced in Part I, we generalize the transformer block and incorporate additional layer types to develop a more versatile architecture. We present a scalable method to efficiently generate an extensive training dataset that significantly exceeds the scope of prior volumes considered in Part I, while still meeting strict accuracy standards. Through our meticulous architecture comparison and comprehensive hyperparameter optimization, we establish that the attention-based architecture performs an order of magnitude better in accuracy than widely adopted NN designs. Finally, we test and apply our emulators to calibrate tension metrics.

astro-ph.CO↗

arXiv2024

Early dark energy constraints with late-time expansion marginalization

Early dark energy (EDE) is an extension to the $Λ$CDM model, proposed to reduce the tension between the measurements of the Hubble constant $H_0$ from the cosmic microwave background (CMB) and from the local cosmic distance ladder. However, this model increases the $S_8$ tension between CMB and large scale structure measurements. Analyses of galaxy clustering and lensing correlation functions report a decreased preference for EDE and its effect on the Hubble tension. Smooth dark energy models affect growth of structure through the background expansion. In this work, we study the inclusion of a general, smooth late-time dark energy modification in combination with EDE and obtain constraints on EDE marginalized over the late-time expansion. We assess the impact on the $S_8$ and Hubble tensions. In order to generalize the late expansion, we use a late dark energy fluid model with a piecewise constant equation of state $w(z)$ over 3, 5 and 10 redshift bins in the window $z \in [0,3]$. We show that, when analyzing ACT and Planck CMB data combined with Pantheon supernovae, BAO from 6dF, SDSS and BOSS, Planck 2018 CMB lensing and Dark Energy Survey cosmic shear and clustering data, the inclusion of a general smooth dark energy modification at late times has no significant effect on $S_8$ and EDE parameter constraints. Using the aforementioned datasets, the EDE fraction constraint with late-time expansion marginalization is $f_\mathrm{EDE} = 0.067^{+0.019}_{-0.027}$ using 3 redshift bins, with similar results for 5 and 10 redshift bins. This work shows that in order to solve simultaneously the Hubble and $S_8$ tensions, one needs a mechanism for increasing the clustering of matter at late times different from a simple change in the background evolution of late dark energy. [Abridged]

astro-ph.CO↗

arXiv2024

Improving Convolutional Neural Networks for Cosmological Fields with Random Permutation

Convolutional Neural Networks (CNNs) have recently been applied to cosmological fields -- weak lensing mass maps and galaxy maps. However, cosmological maps differ in several ways from the vast majority of images that CNNs have been tested on: they are stochastic, typically low signal-to-noise per pixel, and with correlations on all scales. Further, the cosmology goal is a regression problem aimed at inferring posteriors on parameters that must be unbiased. We explore simple CNN architectures and present a novel approach of regularization and data augmentation to improve its performance for lensing mass maps. We find robust improvement by using a mixture of pooling and shuffling of the pixels in the deep layers. The random permutation regularizes the network in the low signal-to-noise regime and effectively augments the existing data. We use simulation-based inference (SBI) to show that the model outperforms CNN designs in the literature. We find a 30% improvement in the constraints of the $S_8$ parameter for simulated Stage-III surveys, including systematic uncertainties such as intrinsic alignments. We explore various statistical errors corresponding to next-generation surveys and find comparable improvements. We expect that our approach will have applications to other cosmological fields as well, such as galaxy maps or 21-cm maps.

astro-ph.CO↗

arXiv2024

Attention-based Neural Network Emulators for Multi-Probe Data Vectors Part I: Forecasting the Growth-Geometry split

We present a new class of machine-learning emulators that accurately model the cosmic shear, galaxy-galaxy lensing, and galaxy clustering real space correlation functions in the context of Rubin Observatory year one simulated data. To illustrate its capabilities in forecasting models beyond the standard $Λ$CDM, we forecast how well LSST Year 1 data will be able to probe the consistency between geometry $Ω^{\rm geo}_\mathrm{m}$ and growth $Ω^{\rm growth}_\mathrm{m}$ dark matter densities in the so-called split $Λ$CDM parameterization. When trained with a few million samples, our emulator shows uniform accuracy across a wide range in an 18-dimensional parameter space. We provide a detailed comparison of three neural network designs, illustrating the importance of adopting state-of-the-art Transformer blocks. Our study also details their performance when computing Bayesian evidence for cosmic shear on three fiducial cosmologies. The transformers-based emulator is always accurate within PolyChord's precision. As an application, we use our emulator to study the degeneracies between dark energy models and growth geometry split parameterizations. We find that the growth-geometry split remains to be a meaningful test of the smooth dark energy assumption.

astro-ph.CO↗

arXiv2023

Constraining Baryonic Physics with DES Y1 and Planck data -- Combining Galaxy Clustering, Weak Lensing, and CMB Lensing

We constrain cosmology and baryonic feedback scenarios with a joint analysis of weak lensing, galaxy clustering, cosmic microwave background (CMB) lensing, and their cross-correlations (so-called 6$\times$2) using data from the Dark Energy Survey (DES) Y1 and the Planck satellite mission. Noteworthy features of our 6$\times$2 pipeline are: We extend CMB lensing cross-correlation measurements to a band surrounding the DES Y1 footprint (a $\sim 25\%$ gain in pairs), and we develop analytic covariance capabilities that account for different footprints and all cross-terms in the 6$\times$2 analysis. We also measure the DES Y1 cosmic shear two-point correlation function (2PCF) down to $0.^\prime 25$, but find that going below $2.^\prime 5$ does not increase cosmological information due to shape noise. We model baryonic physics uncertainties via the amplitude of Principal Components (PCs) derived from a set of hydro-simulations. Given our statistical uncertainties, varying the first PC amplitude $Q_1$ is sufficient to model small-scale cosmic shear 2PCF. For DES Y1+Planck 6$\times$2 we find $S_8=0.799\pm0.016$, comparable to the 5$\times$2 result of DES Y3+SPT/Planck $S_8=0.773\pm0.016$. Combined with our most informative cosmology priors -- baryon acoustic oscillation (BAO), big bang nucleosynthesis (BBN), type Ia supernovae (SNe Ia), and Planck 2018 EE+lowE, we measure $S_8=0.817\pm 0.011$. Regarding baryonic physics constraints, our 6$\times$2 analysis finds $Q_1=2.8\pm1.8$. Combined with the aforementioned priors, it improves the constraint to $Q_1=3.5\pm1.3$. For comparison, the strongest feedback scenario considered in this paper, the cosmo-OWLS AGN ($ΔT_\mathrm{heat}=10^{8.7}$ K), corresponds to $Q_1=5.84$.

astro-ph.CO↗

arXiv2023

Growth and Geometry Split in Light of the DES-Y3 Survey

We test the smooth dark energy paradigm using Dark Energy Survey (DES) Year 1 and Year 3 weak lensing and galaxy clustering data. Within the $Λ$CDM and $w$CDM model we separate the expansion and structure growth history by splitting $Ω_\mathrm{m}$ (and $w$) into two meta-parameters that allow for different evolution of growth and geometry in the Universe. We consider three different combinations of priors on geometry from CMB, SNIa, BAO, BBN that differ in constraining power but have been designed such that the growth information comes solely from the DES weak lensing and galaxy clustering. For the DES-Y1 data we find no detectable tension between growth and geometry meta-parameters in both the $Λ$CDM and $w$CDM parameter space. This statement also holds for DES-Y3 cosmic shear and 3x2pt analyses. For the combination of DES-Y3 galaxy-galaxy lensing and galaxy clustering (2x2pt) we measure a tension between our growth and geometry meta-parameters of 2.6$σ$ in the $Λ$CDM and 4.48$σ$ in the $w$CDM model space, respectively. We attribute this tension to residual systematics in the DES-Y3 RedMagic galaxy sample rather than to new physics. We plan to investigate our findings further using alternative lens samples in DES-Y3 and future weak lensing and galaxy clustering datasets.

astro-ph.CO↗