arXiv subjects

István Csabai

Publications and source records attributed to István Csabai.

At least 19 recordsLinked to original sources

arXiv2026

Million-scale multimodal pollen microscopy with expert-guided foundation models

Automated pollen identification from microscopy remains a bottleneck in aerobiology, palaeoecology and biodiversity monitoring, because scalable systems must generalise across specimen preparation, scanner settings and geographic origins while retaining palynological interpretability. To address this gap, we present a million-scale multimodal pollen microscopy resource, Pollen AI Atlas, assembled from pure-species whole-slide bright-field images spanning four geographic origins, four scanner settings, 45 genera, and one family-only taxon across 31 botanical families. Seeded by one manually selected exemplar per source slide, token-level mining and filtering produced 1,511,390 released grain detections with 99.6\% proposal precision in expert-curated test regions. Each detection was paired with machine-generated grain-level morphological captions from five open-weight vision--language models, guided by expert-verified palynological anchors, yielding structured descriptions of aperture systems, wall ornamentation, shape and size. Among the evaluated models, Gemma4 provided the most controlled primary caption set, combining tight length control, no detected taxon-name or numeric-size leakage and the strongest text-retrieval performance. Baseline benchmarks with frozen visual features reached 88.16\% top-1 accuracy, while cross-regional retrieval showed that caption-derived text embeddings remained robust when image similarity degraded (mAP@20 0.811 versus 0.262). Released data, annotations, captions, splits, code, and weights provide a benchmark for pollen recognition, cross-regional domain adaptation and domain-specific multimodal microscopy learning.

cs.CV↗

arXiv2026

Cylindrical cosmological simulations with StePS

The global topology of the Universe can affect long-range gravitational forces via boundary conditions. Detailed studies of non-trivial topologies require simulations that natively adopt such geometries. Cosmological $N$-body simulations typically evolve matter in a periodic cubic box. While numerically convenient, this imposes a non-trivial three-torus topology that affects long-range gravitational forces, potentially biasing large-scale statistics. We introduce a compactified simulation framework that is only periodic along a single axis, characterised by an infinite topology with isotropic boundary conditions towards the perpendicular directions, namely, a $\mathrm{S}^1\times\mathbb{R}^2$ (slab) topology. This new simulation geometry is ideal for simulating systems with cylindrical symmetries such as filaments or certain anisotropic cosmological models. We compactified the comoving space via an inverse stereographic projection along the radial direction of a periodic cylinder. Then, we evolved the particles based on Newtonian dynamics. A smoothly varying spatial and mass resolution with radius suppresses edge artefacts at the free outer boundary. Our implementation in the StePS (STEreographically Projected cosmological Simulations) framework uses a direct force calculation that maps efficiently to GPUs, as well as an Octree force calculation for use on large CPU clusters. The cylindrical domain's topology enables fully self-consistent simulations to be run in the $\mathrm{S}^1\times\mathbb{R}^2$ manifold, while mitigating any periodic-image artefacts with respect to targets whose symmetries are mismatched to a cubic box. The main trade-off is a radially varying resolution with distinct systematics and analysis requirements. Finally, we demonstrate the accuracy of the new simulation method via a standard lambda cold dark matter cosmological simulation.

astro-ph.CO↗

arXiv2026

STEPSIC: Initial condition generator for stereographic cosmological simulations

Conventional cosmological initial condition generators are designed exclusively for fully periodic cubic domains and cannot produce the non-periodic, observer-centric configurations required by stereographically projected N-body codes such as StePS. We present STEPSIC, an open-source initial condition generator that extends Lagrangian perturbation theory-based initial conditions to the spherical and cylindrical geometries used by StePS, while also supporting cuboid domains with arbitrary aspect ratios. The code constructs Gaussian random density fields on anisotropy-free Fourier grids with cubic voxels, applies first- and second-order LPT to obtain displacement and velocity fields, and interpolates these onto particles via B-spline mass-assignment kernels with Fourier-space deconvolution. For stereographic geometries, a multiresolution scheme maps displacement fields across the radially varying particle mass resolution intrinsic to the projection. Both standard and paired-and-fixed variance-reduced realizations are supported. In periodic cubic boxes, the recovered matter power spectrum agrees with the input linear theory prediction to better than 0.5% up to half the Nyquist wavenumber, independent of box aspect ratio (tested up to 10:1). Cross-validation against monofonic using identical white noise fields yields sub-percent power spectrum agreement, with a small residual offset consistent with differences between two independent implementations. Full N-body evolution of matched cylindrical StePS runs confirms that second-order LPT correctly suppresses the 2-3% transient power excess present in first-order initial conditions.

astro-ph.CO↗

arXiv2025

Integrated Sachs-Wolfe maps from the Gower Street $w$CDM simulations

The late-time linear Integrated Sachs-Wolfe (ISW) effect directly probes the dynamics of cosmic acceleration and the nature of dark energy. Detecting these weak, secondary temperature anisotropy signals of the CMB requires accurate theoretical predictions of their amplitude across cosmological models. By extending the pyGenISW package, previously limited to $Λ$CDM, we aim to generate full-sky ISW maps for a suite of 791 $w$CDM cosmologies using the Gower Street N-body simulations, thereby enabling ISW analyses across a broader dark-energy parameter space. We make our code and ISW data publicly available. We compute the ISW signals by tracing the time evolution of the gravitational potential across large-volume simulations that span dark energy equation of state parameters from phantom to quintessence, $-1.79 \lesssim w \lesssim -0.34$. These data are projected onto the sphere using HEALPix to obtain full-sky temperature maps. We validate our pipeline by comparing the measured ISW angular power spectra and ISW-density cross-correlations against linear theory expectations ($2 \leq \ell \leq 200$) computed with benchmarks from the pyCCL library. The agreement is excellent across the multipole range where the ISW contribution is expected to dominate, confirming the reliability of our modelling of gravitational-potential evolution. With additional tests of the ISW signal's strength in density extrema, as well as comparing all models to a reference $Λ$CDM cosmology, we found that quintessence-like models ($w > -1$) show higher ISW amplitudes than phantom models ($w < -1$), consistent with enhanced late-time decay of gravitational potentials. The consistency of our $w$CDM ISW maps and their agreement with theory predictions confirm the robustness of our methodology, establishing it as a reliable tool for theoretical and observational ISW-LSS analyses.

astro-ph.CO↗

arXiv2025

Constraints on AvERA Cosmologies from Cosmic Chronometers and Type Ia Supernovae

We constrain AvERA cosmologies in comparison with the flat $Λ$CDM model using cosmic chronometer (CC) data and the Pantheon+ sample of type Ia supernovae (SNe Ia). The analysis includes fits to both CC and SN datasets using the \texttt{dynesty} dynamic nested sampling algorithm. For model comparison, we use the Bayesian model evidences and Anderson-Darling tests applied to the normalized residuals to assess consistency with a standard normal distribution. Best-fit parameters are derived within the redshift ranges $z \leq 2$ for CCs and $z \leq 2.3$ for SNe. For the baseline AvERA cosmology, we obtain best-fit values of the Hubble constant of ${H_0=68.32_{-3.27}^{+3.21}~\mathrm{km~s^{-1}~Mpc^{-1}}}$ from the CC analysis and ${H_0=71.99_{-1.03}^{+1.05}~\mathrm{km~s^{-1}~Mpc^{-1}}}$ from the SN analysis, each consistent within $1σ$ with the corresponding AvERA simulation value of $H(z=0)$. While both the CC and SN datasets yield higher Bayesian evidence for the flat $Λ$CDM model, they favor the AvERA cosmologies according to the Anderson-Darling test. We have identified signs of overfitting in each model, which suggests the possibility of overestimating the uncertainties in the Pantheon+ covariance matrix.

astro-ph.CO↗

arXiv2023

Complementary Cosmological Simulations

Cosmic variance limits the accuracy of cosmological N-body simulations, introducing bias in statistics such as the power spectrum, halo mass function, or the cosmic shear. We provide new methods to measure and reduce the effect of cosmic variance in existing and new simulations. We ran pairs of simulations using phase-shifted initial conditions with matching amplitudes. We set the initial amplitudes of the Fourier modes to ensure that the average power spectrum of the pair is equal to the cosmic mean power spectrum from linear theory. The average power spectrum of a pair of such simulations remains consistent with the estimated nonlinear spectra of the state-of-the-art methods even at late times. We also show that the effect of cosmic variance on any analysis involving a cosmological simulation can be estimated using the complementary pair of the original simulation. To demonstrate the effectiveness of our novel technique, we simulated a complementary pair of the original Millennium run and quantified the degree to which cosmic variance affected its the power spectrum. The average power spectrum of the original and complementary Millennium simulation was able to directly resolve the baryon acoustic oscillation features.

astro-ph.CO↗

arXiv2022

Deep Weighted Monte Carlo: A hybrid option pricing framework using neural networks

Recent studies have demonstrated the efficiency of Variational Autoencoders (VAE) to compress high-dimensional implied volatility surfaces into a low dimensional representation. Although this method can be effectively used for pricing vanilla options, it does not provide any explicit information about the dynamics of the underlying asset. In our work we present an effective way to overcome this problem. We use a Weighted Monte Carlo approach to first generate paths from a simple a priori Brownian dynamics, and then calculate path weights to price options correctly. We develop and successfully train a neural network that is able to assign these weights directly from the latent space. Combining the encoder network of the VAE and this new "weight assigner" module, we are able to build a dynamic pricing framework which cleanses the volatility surface from irrelevant noise fluctuations, and then can price not just vanillas, but also exotic options on this idealized vol surface. This pricing method can provide relative value signals for option traders.

q-fin.CP↗

arXiv2022

Photometric redshifts for quasars from WISE-PS1-STRM

Three-dimensional wide-field galaxy surveys are fundamental for cosmological studies. For higher redshifts (z > 1.0), where galaxies are too faint, quasars still trace the large-scale structure of the Universe. Since available telescope time limits spectroscopic surveys, photometric methods are efficient for estimating redshifts for many quasars. Recently, machine learning methods are increasingly successful for quasar photometric redshifts, however, they hinge on the distribution of the training set. Therefore a rigorous estimation of reliability is critical. We extracted optical and infrared photometric data from the cross-matched catalogue of the WISE All-Sky and PS1 3$π$ DR2 sky surveys. We trained an XGBoost regressor and an artificial neural network on the relation between color indices and spectroscopic redshift. We approximated the effective training set coverage with the K nearest neighbors algorithm. We estimated reliable photometric redshifts of 2,879,298 quasars which overlap with the training set in feature space. We validated the derived redshifts with an independent, clustering-based redshift estimation technique. The final catalog is publicly available.

astro-ph.IM↗

arXiv2022

An empirical nonlinear power spectrum overdensity-response

Context. The overdensity inside a cosmological sub-volume and the tidal fields from its surroundings affect the matter distribution of the region. The resulting difference between the local and global power spectra is characterized by the response function. Aims. Our aim is to provide a new, simple, and accurate formula for the power spectrum overdensity response at highly nonlinear scales based on the results of cosmological simulations and paying special attention to the lognormal nature of the density field. Methods. We measured the dark matter power spectrum amplitude as a function of the overdensity ($δ_W$) in $N$-body simulation subsamples. We show that the response follows a power-law form in terms of $(1+δ_W)$, and we provide a new fit in terms of the variance, $σ(L)$, of a sub-volume of size $L$. Results. Our fit has a similar accuracy and a comparable complexity to second-order standard perturbation theory on large scales, but it is also valid for nonlinear (smaller) scales, where perturbation theory needs higher-order terms for a comparable precision. Furthermore, we show that the lognormal nature of the overdensity distribution causes a previously unidentified bias: the power spectrum amplitude for a subsample with an average density is typically underestimated by about $-2σ^2$. Although this bias falls to the sub-percent level above characteristic scales of $200Mpch^{-1}$, taking it into account improves the accuracy of estimating power spectra from zoom-in simulations and smaller high-resolution surveys embedded in larger low-resolution volumes.

astro-ph.CO↗

arXiv2021

Decomposition of stellar populations in CosmoDC2 galaxies using SCARLET and Deep Learning

We are presenting a novel, Deep Learning based approach to estimate the normalized broadband spectral energy distribution (SED) of different stellar populations in synthetic galaxies. In contrast to the non-parametric multiband source separation algorithm, SCARLET - where the SED and morphology are simultaneously fitted - in our study we provide a morphology-independent, statistical determination of the SEDs, where we only use the color distribution of the galaxy. We developed a neural network (sedNN) that accurately predicts the SEDs of the old, red and young, blue stellar populations of realistic synthetic galaxies from the color distribution of the galaxy-related pixels in simulated broadband images. We trained and tested the network on a subset of the recently published CosmoDC2 simulated galaxy catalog containing about 3,600 galaxies. The model performance was compared to the results of SCARLET, where we found that sedNN can predict the SEDs with 4-5% accuracy on average, which is about two times better than applying SCARLET. We also investigated the effect of this improvement on the flux determination accuracy of the bulge and disk. We found that using more accurate SEDs decreases the error in the flux determination of the components by approximately 30%.

astro-ph.IM↗

arXiv2021

GPU-Accelerated Hierarchical Bayesian Inference with Application to Modeling Cosmic Populations: CUDAHM

We describe a computational framework for hierarchical Bayesian inference with simple (typically single-plate) parametric graphical models that uses graphics processing units (GPUs) to accelerate computations, enabling deployment on very large datasets. Its C++ implementation, CUDAHM (CUDA for Hierarchical Models) exploits conditional independence between instances of a plate, facilitating massively parallel exploration of the replication parameter space using the single instruction, multiple data architecture of GPUs. It provides support for constructing Metropolis-within-Gibbs samplers that iterate between GPU-accelerated robust adaptive Metropolis sampling of plate-level parameters conditional on upper-level parameters, and Metropolis-Hastings sampling of upper-level parameters on the host processor conditional on the GPU results. CUDAHM is motivated by demographic problems in astronomy, where density estimation and linear and nonlinear regression problems must be addressed for populations of thousands to millions of objects whose features are measured with possibly complex uncertainties. We describe a thinned latent point process framework for modeling such demographic data. We demonstrate accurate GPU-accelerated parametric conditional density deconvolution for simulated populations of up to 300,000 objects in ~1 hour using a single NVIDIA Tesla K40c GPU. Supplementary material provides details about the CUDAHM API and the demonstration problem.

astro-ph.IM↗

arXiv2021

The anisotropy of the power spectrum in periodic cosmological simulations

The classical gravitational force on a torus is anisotropic and always lower than Newton's $1/r^2$ law. We demonstrate the effects of periodicity in dark matter only $N$-body simulations of spherical collapse and standard $Λ$CDM initial conditions. Periodic boundary conditions cause an overall negative and anisotropic bias in cosmological simulations of cosmic structure formation. The lower amplitude of power spectra of small periodic simulations are a consequence of the missing large scale modes and the equally important smaller periodic forces. The effect is most significant when the largest mildly non-linear scales are comparable to the linear size of the simulation box, as often is the case for high-resolution hydrodynamical simulations. Spherical collapse morphs into a shape similar to an octahedron. The anisotropic growth distorts the large-scale $Λ$CDM dark matter structures. We introduce the direction-dependent power spectrum invariant under the octahedral group of the simulation volume and show that the results break spherical symmetry.

astro-ph.CO↗

arXiv2021

The rich still get richer: Empirical comparison of preferential attachment via linking statistics in Bitcoin and Ethereum

Bitcoin and Ethereum transactions present one of the largest real-world complex networks that are publicly available for study, including a detailed picture of their time evolution. As such, they have received a considerable amount of attention from the network science community, beside analysis from an economic or cryptography perspective. Among these studies, in an analysis on the early instance of the Bitcoin network, we have shown the clear presence of the preferential attachment, or "rich-get-richer" phenomenon. Now, we revisit this question, using a recent version of the Bitcoin network that has grown almost 100-fold since our original analysis. Furthermore, we additionally carry out a comparison with Ethereum, the second most important cryptocurrency. Our results show that preferential attachment continues to be a key factor in the evolution of both the Bitcoin and Ethereum transactoin networks. To facilitate further analysis, we publish a recent version of both transaction networks, and an efficient software implementation that is able to evaluate linking statistics necessary for learn about preferential attachment on networks with several hundred million edges.

physics.soc-ph↗

arXiv2021

The effect of emission lines on the performance of photometric redshift estimation algorithms

We investigate the effect of strong emission line galaxies on the performance of empirical photometric redshift estimation methods. In order to artificially control the contribution of photometric error and emission lines to total flux, we develop a PCA-based stochastic mock catalogue generation technique that allows for generating infinite signal-to-noise ratio model spectra with realistic emission lines on top of theoretical stellar continua. Instead of running the computationally expensive stellar population synthesis and nebular emission codes, our algorithm generates realistic spectra with a statistical approach, and - as an alternative to attempting to constrain the priors on input model parameters - works by matching output observational parameters. Hence, it can be used to match the luminosity, colour, emission line and photometric error distribution of any photometric sample with sufficient flux-calibrated spectroscopic follow-up. We test three simple empirical photometric estimation methods and compare the results with and without photometric noise and strong emission lines. While photometric noise clearly dominates the uncertainty of photometric redshift estimates, the key findings are that emission lines play a significant role in resolving colour space degeneracies and good spectroscopic coverage of the entire colour space is necessary to achieve good results with empirical photo-z methods. Template fitting methods, on the other hand, must use a template set with sufficient variation in emission line strengths and ratios, or even better, first estimate the redshift empirically and fit the colours with templates at the best-fit redshift to calculate the K-correction and various physical parameters.

astro-ph.GA↗

arXiv2020

A common explanation of the Hubble tension and anomalous cold spots in the CMB

The standard cosmological paradigm narrates a reassuring story of a universe currently dominated by an enigmatic dark energy component. Disquietingly, its universal explaining power has recently been challenged by, above all, the $\sim4σ$ tension in the values of the Hubble constant. Another, less studied anomaly is the repeated observation of integrated Sachs-Wolfe imprints $\sim5\times$ stronger than expected in the $Λ$CDM model from R>100 $Mpc/h$ super-structures. Here we show that the inhomogeneous AvERA model of emerging curvature is capable of telling a plausible albeit radically different story that explains both observational anomalies without dark energy. We demonstrate that while stacked imprints of R>100 $Mpc/h$ supervoids in cosmic microwave background temperature maps can discriminate between the AvERA and $Λ$CDM models, their characteristic differences may remain hidden using alternative void definitions and stacking methodologies. Testing the extremes, we then also show that the CMB Cold Spot can plausibly be explained in the AvERA model as an ISW imprint. The coldest spot in the AvERA map is aligned with multiple low-$z$ supervoids with R>100 $Mpc/h$ and central underdensity $δ_{0}\approx-0.3$, resembling the observed large-scale galaxy density field in the Cold Spot area. We hence conclude that the anomalous imprint of supervoids may well be the canary in the coal mine, and existing observational evidence for dark energy should be re-interpreted to further test alternative models.

astro-ph.CO↗

arXiv2019

Galaxy shape measurement with convolutional neural networks

We present our results from training and evaluating a convolutional neural network (CNN) to predict galaxy shapes from wide-field survey images of the first data release of the Dark Energy Survey (DES DR1). We use conventional shape measurements as ground truth from an overlapping, deeper survey with less sky coverage, the Canada-France Hawaii Telescope Lensing Survey (CFHTLenS). We demonstrate that CNN predictions from single band DES images reproduce the results of CFHTLenS at bright magnitudes and show higher correlation with CFHTLenS at fainter magnitudes than maximum likelihood model fitting estimates in the DES Y1 im3shape catalogue. Prediction of shape parameters with a CNN is also extremely fast, it takes only 0.2 milliseconds per galaxy, improving more than 4 orders of magnitudes over forward model fitting. The CNN can also accurately predict shapes when using multiple images of the same galaxy, even in different color bands, with no additional computational overhead. The CNN is again more precise for faint objects, and the advantage of the CNN is more pronounced for blue galaxies than red ones when compared to the DES Y1 metacalibration catalogue, which fits a single Gaussian profile using riz band images. We demonstrate that CNN shape predictions within the metacalibration self-calibrating framework yield shear estimates with negligible multiplicative bias, $ m < 10^{-3}$, and no significant PSF leakage. Our proposed setup is applicable to current and next generation weak lensing surveys where higher quality ground truth shapes can be measured in dedicated deep fields.

astro-ph.CO↗

arXiv2019

StePS: A Multi-GPU Cosmological N-body Code for Compactified Simulations

We present the multi-GPU realization of the StePS (Stereographically Projected Cosmological Simulations) algorithm with MPI-OpenMP-CUDA hybrid parallelization and nearly ideal scale-out to multiple compute nodes. Our new zoom-in cosmological direct N-body simulation method simulates the infinite universe with unprecedented dynamic range for a given amount of memory and, in contrast to traditional periodic simulations, its fundamental geometry and topology match observations. By using a spherical geometry instead of periodic boundary conditions, and gradually decreasing the mass resolution with radius, our code is capable of running simulations with a few gigaparsecs in diameter and with a mass resolution of $\sim 10^{9}M_{\odot}$ in the center in four days on three compute nodes with four GTX 1080Ti GPUs in each. The code can also be used to run extremely fast simulations with reasonable resolution for fitting cosmological parameters. These simulations are useful for prediction needs of large surveys. The StePS code is publicly available for the research community.

astro-ph.CO↗

arXiv2019

Weak lensing cosmology with convolutional neural networks on noisy data

Weak gravitational lensing is one of the most promising cosmological probes of the late universe. Several large ongoing (DES, KiDS, HSC) and planned (LSST, EUCLID, WFIRST) astronomical surveys attempt to collect even deeper and larger scale data on weak lensing. Due to gravitational collapse, the distribution of dark matter is non-Gaussian on small scales. However, observations are typically evaluated through the two-point correlation function of galaxy shear, which does not capture non-Gaussian features of the lensing maps. Previous studies attempted to extract non-Gaussian information from weak lensing observations through several higher-order statistics such as the three-point correlation function, peak counts or Minkowski-functionals. Deep convolutional neural networks (CNN) emerged in the field of computer vision with tremendous success, and they offer a new and very promising framework to extract information from 2 or 3-dimensional astronomical data sets, confirmed by recent studies on weak lensing. We show that a CNN is able to yield significantly stricter constraints of ($σ_8, Ω_m$) cosmological parameters than the power spectrum using convergence maps generated by full N-body simulations and ray-tracing, at angular scales and shape noise levels relevant for future observations. In a scenario mimicking LSST or Euclid, the CNN yields 2.4-2.8 times smaller credible contours than the power spectrum, and 3.5-4.2 times smaller at noise levels corresponding to a deep space survey such as WFIRST. We also show that at shape noise levels achievable in future space surveys the CNN yields 1.4-2.1 times smaller contours than peak counts, a higher-order statistic capable of extracting non-Gaussian information from weak lensing maps.

astro-ph.CO↗