arXiv subjects

Benedikt Schosser

Publications and source records attributed to Benedikt Schosser.

8 recordsLinked to original sources

arXiv2026

The Latent Information Geometry of Jet Classification

Latent representations are an important theme in modern machine learning. Any network training with the notion of locality introduces a latent geometry which we can analyze with the help of differential geometry, specifically information geometry. We introduce the main concepts needed to analyze learned latent geometries, specifically curvature and nonmetricities, and show how they can be used for decoder and classifier geometries. We then apply our new methods to understand the physics behind binary quark-gluon classification and three-fold fat jet tagging.

hep-ph↗

arXiv2025

Markov Walk Exploration of Model Spaces: Bayesian Selection of Dark Energy Models with Supernovae

Central to model selection is a trade-off between performing a good fit and low model complexity: A model of higher complexity should only be favoured over a simpler model if it provides significantly better fits. In Bayesian terms, this can be achieved by considering the evidence ratio, enabling choices between two competing models. We generalise this concept by constructing Markovian random walks for exploring the entire model space. In analogy to the logarithmic likelihood ratio in parameter estimation problem, the process is governed by the logarithmic evidence ratio. We apply our methodology to selecting a polynomial for the dark energy equation of state function $w(a)$ on the basis of data for the supernova distance-redshift relation.

astro-ph.CO↗

arXiv2025

Direct reconstruction of the Reionization history from 21cm 2D Power Spectra

The 21cm line from the spin-flip transition of neutral hydrogen (HI) provides a unique window into the Epoch of Reionization (EoR), the final phase transition of our Universe. The Square Kilometre Array (SKA) enables precise measurements of 21cm fluctuations that trace ionization, temperature, and density fluctuations of the intergalactic medium (IGM). Nevertheless, a direct reconstruction of the timeline of the EoR in terms of the progress of ionization remains an ongoing challenge due to the highly non-Gaussian nature and thus intractable likelihood of the 21cm signal. Here, we present EoRFlow, a simulation-based inference (SBI) framework for reconstructing the global neutral hydrogen fraction $x_{\mathrm{HI}}(z)$ directly from 2D cylindrically averaged power spectra (2DPS) of the 21cm signal. We validate our method on realistic mock datasets for SKA-Low. Bypassing the need for explicit likelihood formulations, our approach enables fast, unbiased posterior estimation of the $x_{\mathrm{HI}}$ evolution in narrow redshift slices, allowing for piecewise reconstruction of the global reionization history. By directly inferring the reionization history from 21cm power spectra, our framework provides a scalable and robust path forward for 21cm cosmology in the SKA era.

astro-ph.CO↗

arXiv2025

Starobinsky in Stereo: SKA-CMB Synergy in SBI

Modern machine learning techniques can unlock the vast cosmological information encoded in forthcoming Square Kilometre Array (SKA) observations. We show that tomographic 21 cm data from the reionisation era can yield stringent tests of inflationary models - here illustrated with Starobinsky $R+R^2$ inflation. Using a simulation-based inference (SBI) framework, we compare neural summaries (convolutional network and vision transformer) with a traditional power spectrum summary and perform a fully joint SBI analysis combining 21 cm data with data of the cosmic microwave background (CMB). Forecasts based on realistic mock observations indicate that SKA alone will achieve constraints competitive with Planck, and that the combined SKA + CMB dataset will tighten bounds on both inflationary and $Λ\mathrm{CDM}$ parameters considerably while improving precision on key astrophysical quantities.

astro-ph.CO↗

arXiv2025

On the full non-Gaussian Surprise statistic and the cosmological concordance between DESI, SDSS and Pantheon+

With the increasing precision of recent cosmological surveys and the discovery of important tensions within the $Λ$CDM paradigm, it is becoming more and more important to develop tools to quantify accurately the discordance between different probes. One such tool is the Surprise statistic, a measure based on the Kullback-Leibler divergence. The Surprise, however, has been up to now applied only under its Gaussian approximation, which can fail to properly capture discordance in cases that deviate significantly from Gaussianity. In this paper we developed the \texttt{klsurprise} code which computes the full numerical non-Gaussian Surprise, and analyse the Surprise for BAO + BBN and supernova data. We test different cosmological models, some of which the parameters deviate significantly from Gaussianity. We find that the non-Gaussianities, mainly present in the Supernova dataset, change the Surprise values significantly from its Gaussian approximation, and reveal a borderline $2.0σ$ tension in the curved $w$CDM model (o$w$CDM) between the combined Pantheon+ and SH0ES (Pantheon+ & SH0ES) data and the dataset which combines SDSS, BOSS and eBOSS BAO. This modest tension is hidden in the Gaussian Surprise approximation. For DESI, the discrepancy with Pantheon+ & SH0ES is at the $1.5σ$ level for o$w$CDM, but a large $3.4σ$ for $Λ$CDM. Removing SH0ES data drops the $Λ$CDM significance to $2.6σ$.

astro-ph.CO↗

arXiv2025

Optimal, fast, and robust inference of reionization-era cosmology with the 21cmPIE-INN

Modern machine learning will allow for simulation-based inference from reionization-era 21cm observations at the Square Kilometre Array. Our framework combines a convolutional summary network and a conditional invertible network through a physics-inspired latent representation. It allows for an efficient and extremely fast determination of the posteriors of astrophysical and cosmological parameters, jointly with well-calibrated and on average unbiased summaries. The sensitivity to non-Gaussian information makes our method a promising alternative to the established power spectra.

astro-ph.CO↗

arXiv2025

Bayesian distances for quantifying tensions in cosmological inference and the surprise statistic

Tensions between cosmological parameters derived through different channels can be a genuine signature of new physics that $Λ$CDM as the standard model is not able to reproduce, in particular in the missing consistency between parameter estimates from measurements the early and late Universe. Or, they could be caused by yet to be understood systematics in the measurements as a more mundane explanation. Commonly, cosmological tensions are stated in terms of mismatches of the posterior parameter distributions, often assuming Gaussian statistics. More importantly, though, would be a quantification if two data sets are consistent to each other before combining them into a joint measurement, ideally isolating hints at individual data points that have a strong influence in generating the tension. For this purpose, we start with statistical divergences applied to posterior distributions following from different data sets and develop the theory of a Fisher metric between two data sets, in analogy to the Fisher metric for different parameter choices. As a topical example, we consider the tension in the Hubble-Lemaître constant $H_0$ from supernova and measurements of the cosmic microwave background, derive a ranking of data points in order of their influence on the tension on $H_0$. For this particular example, we compute Bayesian distance measures and show that in the light of CMB data, supernovae are commonly too bright, whereas the low-$\ell$ CMB spectrum is too high, in agreement with intuition about the parameter sensitivity.

astro-ph.CO↗

arXiv2023

Cornering Extended Starobinsky Inflation with CMB and SKA

Starobinsky inflation is an attractive, fundamental model to explain the Planck measurements, and its higher-order extension may allow us to probe quantum gravity effects. We show that future CMB data combined with the 21cm intensity map from SKA will meaningfully probe such an extended Starobinsky model. A combined analysis will provide a precise measurement and intriguing insight into inflationary dynamics, even accounting for correlations with astrophysical parameters.

astro-ph.CO↗