OPENING REMARKS Room: PI/4-405 - Bob Room

Generative Solutions for Cosmic Problems

• Carolina Cuesta-Lazaro (Flatiron Institute)

Room: PI/4-405 - Bob Room

Dynamic SBI for cosmological field-level inference

• Oleg Savchenko (GRAPPA Institute, University of Amsterdam)

Room: PI/4-405 - Bob Room Simulation-based inference (SBI) enables Bayesian analysis of complex cosmological data when only a forward model is available, while field-level inference (FLI) aims to perform inference in a maximally efficient way and retain more information than summary-statistic pipelines. In this talk, I will highlight recent advances and applications of SBI and FLI in cosmology. First, I will show how field-level SBI can be used to reconstruct cosmological fields from incomplete and noisy data. Using Gaussian neural posterior estimation with a trainable mean and covariance, and combining classical conjugate-gradient solvers with neural networks, our method captures complex spatial correlations, denoises observations, and probabilistically reconstructs missing regions. We demonstrate this approach on the challenging task of inferring the 3D dark matter field and its initial conditions. Finally, I will describe how this method can be embedded in an active-learning framework for dynamic SBI, enabling joint inference of fields and cosmological parameters by steering simulations toward the most relevant regions of parameter space.

Bridging Simulators with Conditional Optimal Transport

• Justine Zeghal (Université de Montréal, Mila)

Room: PI/4-405 - Bob Room Weak lensing convergence maps are expected to be significantly non-Gaussian on small scales, causing the power spectrum to fail as a sufficient statistic for cosmological parameters by discarding information encoded in higher-order correlations. This limitation is particularly pressing in the context of next-generation surveys, which will improve the signal-to-noise ratio and grant access to deeply non-linear scales. To tighten constraints on cosmological parameters, neural-network-based full-field inference methods have recently emerged as a promising avenue; however, they require large suites of computationally expensive simulations. To address this challenge, we propose a new approach to building pixel-level emulators (Zeghal et al., 2025). By learning a minimal conditional transformation between cheap and costly simulations using Conditional Optimal Transport Flow Matching (COT-FM, Kerrigan et al., 2024), we show that we can generate new high-fidelity simulations whose statistics match those of the expensive simulator across cosmological parameters. We validate this by performing full-field inference on the emulated simulations and demonstrate that the resulting posteriors are in excellent agreement with those obtained from the true costly simulations. Notably, our emulator can operate on unpaired datasets. This flexibility enabled us to secure second place in the Weak Lensing Uncertainty Challenge at NeurIPS 2025, where only the simulations and their corresponding cosmological parameters were provided, without access to the initial conditions needed to pair them with their cheap counterparts.

Sensitivity Analysis of the SimBIG-II Forward Model

• Madeline Casas (The University of Texas at Austin)

Room: PI/4-405 - Bob Room Simulation-based inference (SBI) has recently emerged as a promising avenue for inferring cosmological parameters from the 3D distribution of galaxies from large spectroscopic surveys (DESI, PFS, DESI-II). The first iteration of the SimBIG project demonstrated a proof-of-concept for this approach, using a forward-modeling pipeline built from high-fidelity QUIJOTE N-body simulations applied to a subset of the BOSS survey. The initial SimBIG results were promising, producing $H_0$ and $S_8$ constraints up to 1.5 and 1.9x tighter than traditional power spectra analyses. However, scaling this approach to larger survey volumes presents a major challenge, as the computational costs of high-resolution N-body simulations quickly becomes prohibitive. In this work, we present a scalable SimBIG-II approach. We leverage relatively inexpensive simulations from augmented Lagrangian perturbation theory (ALPT) with a highly flexible local and non-local bias prescription. We combine these simulations with a probabilistic emulator so that they retain the predictive power of N-body + HOD simulations. I will present the preliminary results of our sensitivity analysis of the SimBIG-II pipeline. Then, I will discuss the implications for our initial application to the full BOSS survey volume, as well as the subsequent SimBIG-II analyses of DESI galaxy samples.

Strong Gravitational Lensing in the Era of Data-Driven Algorithms

• Yashar Hezaveh

Room: PI/4-405 - Bob Room

Detecting Modeling Bias with Continuous Time Flow Models on Weak Lensing Maps

• Kangning Diao (UC Berkeley)

Room: PI/4-405 - Bob Room Simulation-based inference (SBI) provides a powerful framework for extracting rich information from nonlinear scales in current and upcoming cosmological surveys, and ensuring its robustness requires stringent validation of forward models. In this work, we recast forward model validation as an out-of-distribution (OoD) detection problem within the framework of machine learning (ML)-based SBI. We employ probability density as the metric for OoD detection, and compare various density estimation techniques, demonstrating that field-level probability density estimation via continuous time flow models (CTFM) significantly outperforms feature-level approaches that combine scattering transform (ST) or convolutional neural networks (CNN) with normalizing flows (NFs), as well as NF-based field-level estimators, as quantified by the area under the receiver operating characteristic curve (AUROC). Our analysis shows that CTFM not only excels in detecting OoD samples but also provides a robust metric for model selection. Additionally, we verified CTFM maintains consistent efficacy across different cosmologies while mitigating the inductive biases inherent in NF architectures. Although our proof-of-concept study employs simplified forward modeling and noise settings, our framework establishes a promising pathway for identifying unknown systematics in the cosmology datasets.

Joint inference of mass-maps and cosmology from weak lensing cosmic shear with diffusion models

• Benjamin Remy (The University of Chicago)

Room: PI/4-405 - Bob Room Upcoming Stage-IV galaxy surveys will map the large-scale structure of the Universe with unprecedented precision, requiring analysis methods that can exploit information beyond traditional two-point statistics. Field-level inference offers a principled path forward by working directly with the observed fields, capturing non-Gaussian signatures that summary statistics discard. However, existing approaches typically address either cosmological parameter estimation or field reconstruction in isolation, or rely on explicit inference frameworks that require differentiable forward models and costly MCMC sampling. We present a diffusion-model-based method that performs joint inference of weak lensing convergence maps and cosmological parameters in a single, unified framework. Built on a pixel-space vision transformer, our model learns the joint posterior distribution over both the convergence field and cosmological parameters, conditioned on noisy shear observations. By operating within the implicit inference paradigm, our approach bypasses the need for a differentiable forward model, opening the door to arbitrarily complex simulators. We demonstrate our method on simulated LSST Year-10 weak lensing data generated with log-normal convergence fields in a wCDM cosmology, showing that our approach can recover accurate joint posteriors over both the mass map and cosmological parameters from a single amortized model, validated against MCMC baselines and coverage diagnostics.

Enabling KARMMA as a Tool of Precision Cosmology

• Ivan Espinoza Bustamante (The University of Arizona)

Room: PI/4-405 - Bob Room We will present KARMMA, a field-level inference designed to enable joint inference of cosmology and the convergence field from cosmic shear data. KARMMA is a full-sky Bayesian algorithm that forward-models the convergence field as an augmented lognormal realization. Using N-body simulations, we generate mock cosmic shear observations and use these to validate the cosmological inferences from KARMMA. We find field-level inference with LSST Y1-like data has the potential to improve constraints on the dark energy equation of state by a factor of six relative to a standard power-spectrum approach.

Connecting field-level inference with summary statistics

• Fabian Schmidt (Max Planck Institute for Astrophysics)

Room: PI/4-405 - Bob Room I will discuss how field-level inference can be connected with summary statistics in the context of a perturbative forward model, and what this implies for the information gain obtainable at the field level. I will also present recent quantitative results on the information gain.

Toward Simulation-Based Inference of Inflation with Lattice Simulations

• Drew Jamieson (Max Planck Institute for Astrophysics)

Room: PI/4-405 - Bob Room Simulating inflation is emerging as a powerful technique for studying inflationary phenomenology beyond the standard paradigm of single-field, slow-roll, and perturbative dynamics. In this talk, I will present recent results from lattice simulations of axion-gauge inflationary models that exhibit a rich phenomenology, including an enhanced small-scale power spectrum, a blue-tilted equilateral-peaked, but ultimately non-separable, bispectrum, and higher-order correlation functions that violate parity. I will also present new results characterizing the CMB power spectrum and bispectrum signatures of these models. The output of our simulations consists of realizations of the primordial curvature field imbued with this nontrivial hierarchy of N-point statistics. Crucially, such realizations cannot be generated with standard techniques for primordial non-Gaussian initial conditions, making our approach essential for comprehensive analysis using field-level and simulation-based inference. Using the output of our inflation simulations as initial conditions for N-body simulations, we effectively simulate the entire history of the universe from inflation to late-time large-scale structure, characterizing the observable predictions of axion-gauge inflation for galaxy surveys. These results establish simulating inflation as an indispensable tool for broadening the space of testable early-universe theories.

Field-Level Diffusion Emulators for SBI

• Amanda Lue (Columbia University)

Room: PI/4-405 - Bob Room Cosmological studies with surveys such as DESI, Roman and Euclid will be most powerful if they can exploit the rich information on small, nonlinear scales, which are often removed by conservative cuts due to the difficulty of robustly modelling baryonic physics. We present an accelerated forward-modelling framework aimed at enabling simulation based inference (SBI) while marginalizing over baryonic modelling uncertainties. Or approach learns the mapping from dark matter structure to galaxies directly from hydrodynamical simulations, which provide the most physically complete models of structure formation and feedback currently available. Leveraging simulation suites such as CAMELS that span cosmological and astrophysical parameters as well as multiple subgrid physics models, we seek to generate diverse mock realisations that support cosmological inference while accounting for baryonic systematics. We train a diffusion-based generative model on paired N-body and hydrodynamical simulations to rapidly produce galaxy count fields, with the broader framework designed to extend to realistic galaxy catalogs that include the observables needed for survey-style selection, including broad-band photometry in relevant filters. Although current hydrodynamical training volumes are smaller than survey volumes, the trained model is fully convolutional and can be applied efficiently to much larger volumes needed for analyses and covariance estimation, establishing a promising forward-modelling foundation for SBI with observations.

A field-level emulator for extra fundamental forces

• Daniela Saadeh (Durham University)

Room: PI/4-405 - Bob Room We present a field-level framework to emulate the effects of extra fundamental forces on the cosmic web. This approach is designed to enable field-level inference with data from Stage IV cosmological surveys. Building on the reaction method, which models the nonlinear matter power spectrum in modified gravity as corrections to a "pseudo'' $\Lambda$CDM cosmology, we extend the method to full field-level predictions by applying it to the output of N-body simulations, including both positions and velocities. Our neural network predicts the field-level correction ("reaction'') to a pseudo$\Lambda$CDM simulation whose linear clustering matches that of the target. The emulator achieves sub-percent accuracy across a broad range of summary statistics, including 0.4% agreement in the matter power spectrum at scales k < 1 Mpc/h, and 2% accuracy in redshift-space distortion multipoles at k < 0.3 Mpc/h. We also validate the emulator against N-body simulations with increased force resolution and time steps, confirming the robustness of its performance. These results demonstrate that our framework is a practical and reliable tool for incorporating screened modified gravity models into field-level cosmological inference, enabling stringent tests of extra fundamental forces at cosmological scales.

The Manticore Project: Field-Level Posterior Inference as a Laboratory for Cosmology and Galaxy Formation

• Jens Jasche

Room: PI/4-405 - Bob Room Traditional cosmological analyses describe ensemble-averaged properties of structure formation. Bayesian field-level inference offers a fundamentally different capability: by constructing posterior ensembles of initial conditions directly constrained by galaxy survey data, one recovers physically consistent realizations of the specific Universe we inhabit. This eliminates cosmic variance as a systematic and turns individual structures into controlled tests of physical models, a mode of inference unavailable to any statistical approach. I will present the Manticore project as a large-scale demonstration of this capability, spanning from methodology to concrete scientific consequences. The Manticore project applies this framework to two complementary datasets. Constrained by the all-sky 2MASS galaxy catalogue, the local reconstruction recovers the three-dimensional matter and velocity distribution within 200 Mpc, achieving the highest Bayesian evidence for the peculiar velocity field across five independent datasets and robustly identifying fourteen nearby galaxy clusters. Constrained by the combined SDSS and BOSS spectroscopic surveys, the reconstruction extends to a (4 h⁻¹ Gpc)³ volume at ∼4 h⁻¹ Mpc resolution, validated against observations withheld from the inference: a cross-correlation with Planck CMB lensing and a kinetic SZ detection from velocity-weighted cluster stacking. Having accurate, uncertainty-quantified models of the local density and velocity fields opens new avenues for precision cosmology. Peculiar velocity corrections are a dominant systematic in local H₀ measurements; replacing standard linear reconstructions with the Manticore velocity posterior in a SN-free Cepheid distance ladder analysis yields H₀ = 71.1 ± 1.4 km/s/Mpc with an 18% uncertainty reduction. The reconstructed density field further reveals that SN Ia rates in nearby superclusters exceed expectations from matter overdensities alone by factors of two to five, indicating that the local large-scale structure introduces correlated systematics into SN cosmology samples that standard analyses do not capture. Finally, constrained hydrodynamical simulations of individual nearby galaxy clusters, confronted against X-ray and SZ observations, expose failures in AGN feedback models that population-level statistics cannot detect. Together, these results illustrate what becomes possible when field-level inference is pushed beyond parameter estimation toward direct confrontation with the observed Universe, and outline an emerging programme for jointly constraining cosmology and astrophysical processes within a single coherent reconstruction.

Field-level inference of the Local Group and its surrounding mass distribution

• Ewoud Wempe (CNRS / LIRA, Observatoire de Paris)

Room: PI/4-405 - Bob Room I will present a Bayesian hierarchical field-level inference framework for the Local Group and its immediate cosmological environment. ΛCDM initial conditions are conditioned on observational constraints on the masses, relative position, and velocities of the Milky Way and M31 haloes, and on the surrounding velocity field traced by isolated galaxies within 5 Mpc. This yields statistically representative realizations of the local matter distribution that simultaneously reproduce the Local Group system and its surrounding flow. I will describe this multi-resolution inference framework and show that, within a standard ΛCDM cosmology, these data require a strongly flattened mass distribution around the Local Group, aligned with the Local Sheet and bordered by underdense regions above and below the plane. In this geometry, the quiet Hubble flow with respect to the Local Group is consistent with the dynamical mass of the Milky Way-M31 system, resolving a long-standing tension in spherical models and showing that light traces mass in our neighbourhood on these scales.

Type Ia Supernova Peculiar Velocity Inference with the BORG Forward Model Towards ZTF Applications

• Mahmoud Osman (LPNHE/IN2P3/CNRS)

Room: PI/4-405 - Bob Room Peculiar velocities of matter tracers, arising from gravitational infall into large-scale structures, can be used to determine the growth rate of cosmic structure, $f\sigma_8$,providing a direct test of General Relativity. Measurements of velocity fields therefore constitute an important probe of the standard cosmological model. In this talk, I will present our methodology based on the *Bayesian Origin Reconstruction from Galaxies* (BORG) framework, a probabilistic forward-modeling approach designed to infer both the initial conditions and the late-time large-scale structures of the Universe. This framework jointly infers the density and peculiar velocity fields. We explore the application of this approach to Type Ia supernovae in order to fully exploit the rich low-redshift ($z<0.1$) supernova dataset provided by the Zwicky Transient Facility (ZTF). Accurate measurements of $f\sigma_8$ at low redshift are particularly informative for tests of gravity, as this is the regime where the effects of cosmic expansion are least dominant. I will present growth rate results from mock catalogues generated within the BORG framework and by the ZTF collaboration. BORG-generated mocks demonstrate accurate recovery of the underlying velocity field and cosmological parameters, while applications to ZTF mock catalogues represent a first step toward a full field-level analysis of peculiar velocities with real supernova samples.

Lyman-alpha forest and high-redshift galaxies at the field level

• Mikhail Ivanov

Room: PI/4-405 - Bob Room I will overview recent advances in modeling of the high-redshift tracers, such as the Lyman-alpha forest, Lyman-alpha emitting galaxies, and Lyman-break galaxies at the field level using large-scale structure effective field theory. I will then discuss various new opportunities that this modeling offers, including generation of high-quality mock catalogs and simulation-based priors for analyses of DESI-II and Spec-S5 data.

FLISBI for LSSCMB

• Adrian Bayer

Room: PI/4-405 - Bob Room

Coverage is not enough: What SBI posteriors of f_NL are actually telling you

• Toka Alokda (Argelander Institute for Astronomy, University of Bonn)

Room: PI/4-405 - Bob Room Simulation-based inference (SBI) is increasingly used to extract cosmological information from complex observables, with reliability typically validated through coverage-based diagnostics such as simulation-based calibration (SBC) and the coverage test of accuracy with random points (TARP). These tests check whether posteriors contain the true parameter value with the expected frequency; a necessary condition for accuracy. However, coverage is insensitive to posterior shapes, so an estimator can pass such tests while exhibiting systematic tail biases or realization-level discrepancies that go undetected. In this talk I present our work where we systematically compare posteriors obtained through likelihood-based inference (LBI) and SBI with contrastive neural ratio estimation (CNRE) for constraining local primordial non-Gaussianity ($f_{\rm NL}^{\rm local}$) from dark matter halo statistics in the Quijote-PNG simulations. Using the power spectrum ($P$), bispectrum ($B$), and wavelet scattering transform (WST) coefficients, we compare posterior distributions across 1000 test realizations, examining higher-order moments, credible interval shapes, and tail behavior. We also use the Wasserstein-2 distance as a shape-sensitive diagnostic to capture discrepancies that aren't captured by other metrics. We show that the $P+B$ SBI posterior is systematically under-confident relative to that from LBI, a miscalibration invisible to standard diagnostics, demonstrating concretely that passing coverage tests does not guarantee posterior faithfulness. We also find that WST coefficients improve $f_{\rm NL}^{\rm local}$ constraints beyond $P+B$ even at large scales ($k_{\rm max} = 0.14\, h\, {\rm Mpc}^{-1}$), supporting field-level summaries as probes of primordial physics.

Pushing field level inference into the non-linear regime with multi-probe analysis

• Benjamin Horowitz (Kavli IPMU)

Room: PI/4-405 - Bob Room Understanding the complex interplay of gas and dark matter is critical for the next generation of cosmological surveys and related inference. While many techniques seek to marginalize over these uncertainties, there is a wealth of information from across astrophysics than can better constrain properties like galaxy formation and evolution. In this talk, I will introduce recent developments in pushing field level inference into these nonlinear scales. These efforts include differentiable models for full dark-matter dynamics (i.e. TreePM), hydrodynamical physics, subgrid physics, and halos/group finding. Not only can differentiability be used for explicit likelihood analysis, but also to greatly accelerate existing simulation-based pipelines. Time permitting, I will also discuss how these techniques will be used for an integrated analysis in the ongoing Prime Focus Spectrograph survey, which will study galaxy evolution from z ~ 7.2 to the present day.

Field-Level Inference of Primordial Non-Gaussianity

• Hugo Simon (CEA Paris-Saclay)

Room: PI/4-405 - Bob Room Field-Level Inference (FLI) needs to be made more tractable and robust at survey scale. To this purpose, I developed fast, differentiable cosmological simulators and introduced a standardized benchmark, showing how to reduce the required model evaluations by orders of magnitude. Building on this, I am working toward FLI constraints on local Primordial Non-Gaussianity (PNG) from DESI. I will show the validation of the pipeline on N-body simulations with HOD-populated galaxies, carefully assessing model fidelity and calibration, progressively incorporating survey realism at the field-level, and in parallel extending the analysis to a multi-probe framework combining galaxy clustering with CMB lensing.

From Perturbative Modeling to Simulation-Based Inference: SimBIG from Stage-III to Stage-IV Galaxy Surveys

• Azadeh Moradinezhad (CNRS - LAPTh)

Room: PI/4-405 - Bob Room

Learning the Universe with CAMELS-SAM: Joint Simulation-Based Inference of Cosmology and Astrophysics on SDSS Galaxy Clustering

• Lucia Perez (Flatiron Institute)

Room: PI/4-405 - Bob Room As part of the Learning the Universe collaboration, we have pushed the CAMELS-SAM simulation suite into observational photometric space, and present initial results for constraints on cosmological and astrophysical parameters for galaxy population statistics for the SDSS Main Galaxy Sample. CAMELS-SAM already included more than 1000 N-body simulations of (100 cMpc)$^3$, covering a vast cosmological parameter space in $\Omega_M$ and $\sigma_8$. Our new and unique forward-model combines: the CAMELS-SAM simulations and framework with the Santa Cruz semi-analytic model for galaxy formation (varying several new parameters for galaxy formation processes); the powerful and flexible Synthesizer software to model synthetic astrophysical observables for simulated galaxies; and physically-motivated analyses for galaxy dust attenuation. With these components, we test how clustering statistics and other galaxy population statistics perform in a simulation-based inference pipeline directly in the observed lightcone. We present initial results and lessons learned in this ambitious application of SBI and galaxy forward modeling, and discuss how it can be scaled up for the next generation of galaxy surveys. Finally, we detail the value of the expanded CAMELS-SAM simulation suite for the testing and development of new field-level methods for galaxy surveys.

Simulation-based cosmological inference from 3D maps of multiple tracers

• Giulio Scelfo (SISSA)

Room: PI/4-405 - Bob Room Simulation-Based Inference (SBI) overcomes problems of likelihood-based estimators allowing for the extraction of information at full-field level. I present a proof-of-concept SBI pipeline to marginally constrain the cosmological parameters $\{\Omega_m, \sigma_8\}$ from large-scale structure observables. Our approach combines fast dark matter simulations with neural emulators that generate galaxy and HI maps. We perform inference both on the power spectrum as a summary statistics and directly on field-level representations of the data, using either one single probe or multiple fields at the same time, while marginalising over baryonic nuisance parameters. We assess systematically the impact of data compression and multi-tracers information on cosmological constraints. We find that multi-probe analyses, combining galaxy and HI fields, improve constraints with respect to single-probe cases. Moving from summary statistics to field-level inference leads to a significant gain in constraining power, with the full 3D approach providing the most accurate and well-calibrated posteriors. When marginalizing over astrophysical effects, field-level methods retain substantially more cosmological information than power-spectrum-based analyses. These results highlight the potential of combining multi-probes observations and full field information into SBI to fully exploit the information content of forthcoming surveys, and provide a baseline for more realistic observational data-based analyses.

Discussion: Precise and accurate inference for SBI and FLI Room: PI/4-405 - Bob Room

Field level inference for LSS and CMB

• Uros Seljak

Room: PI/4-405 - Bob Room I will present recent results of field level inference applied to three different cosmological probes. In the case of gravitational lensing of cosmic microwave background, explicit field level inference can extract both the lensing potential and delensed CMB. Recent developments in MCMC sampling have enabled full hierarchical Bayesian analysis. For galaxy clustering and Baryonic Acoustic Oscillations, both explicit and implicit field level inference are being developed, and I will present recent results and future prospects. Finally, I will present results from implicit field level inference applied to galaxy shear data. In all cases, field level inference leads to significant improvements over the current state of the art analyses.

Probing Cosmology through Higher-Order CMB Lensing Statistics

• Shu-Fan Chen (Columbia University)

Room: PI/4-405 - Bob Room We investigate the cosmological information content of higher-order statistics of the CMB lensing convergence field for near-term experiments similar to the Simons Observatory. Using a field-level forward-modeling pipeline based on ray-traced $N$-body simulations with realistic SO-like lensing reconstruction, we measure non-Gaussian statistics such as Minkowski functionals, peak/minima counts, and train emulators to model their dependence on $\Omega_m$, $A_s$, and $M_\nu$. We quantify the information gain beyond the lensing power spectrum and identify which statistics are most robust to reconstruction noise, highlighting the potential of non-Gaussian statistics to enhance cosmological constraints from upcoming CMB surveys.

Disentangling Feedback and Variance in 1,024 Milky Way-Mass DREAMS Simulations

• Jonah Rose (Princeton Univeristy)

Room: PI/4-405 - Bob Room We introduce a novel framework for simulation-based inference using the DREAMS Project, a suite of 1,024 cosmological hydrodynamical zoom-in simulations of Milky Way-mass halos. This suite is designed to systematically disentangle theoretical uncertainties in galaxy formation physics from intrinsic halo-to-halo variance by varying key astrophysical parameters governing supernova wind energy, wind speed, and AGN feedback efficiency within the IllustrisTNG model [arXiv:2512.00148]. To overcome the computational bottleneck of evaluating this high-dimensional parameter space, we utilize a hierarchical generative machine learning framework. By incorporating conditional normalizing flows and Variational Diffusion Models, we accurately emulate both central host properties and variable-length satellite populations [arXiv:2409.02980]. We then introduce a novel observational weighting scheme constrained by the empirical stellar mass-halo mass relation [arXiv:2602.03613]. This approach yields pseudo-posterior constraints that reveal broad degeneracies in fiducial feedback parameters, demonstrating that standard single-model tuning misses complex parameter interdependencies. Applying this inference framework allows us to robustly assess the impact of feedback variations versus accretion history. For central galaxies, we demonstrate that specific structural shifts are driven by specific merger histories, such as the Gaia-Sausage-Enceladus event, though immense halo-to-halo scatter persists. For satellites, we show that intrinsic variance overwhelmingly dominates population statistics, while identifying a persistent tension regarding extended half-light radii observed in the SAGA survey [arXiv:2512.02095]. Finally, we outline ongoing extensions of this inference framework to multi-code simulations (FIRE3, RAMSES, ChaNGa) along with new simulation designs that incorporate mass-varied and resolution-varied suites spanning $10^9$ to $10^{14}$ solar masses; establishing a new method to understand parameter and resolution variations across a full range of halo masses.

What Dominates the Uncertainty on Local Dark Matter Speed Distributions?

• Ethan Lilie (Princeton University)

Room: PI/4-405 - Bob Room Dark matter direct detection experiments require information about the local dark matter speed distribution to produce constraints on dark matter candidates, or infer their properties in the event of a discovery. I will discuss how the uncertainty in the dark matter speed distribution near the Sun is affected by baryonic feedback, halo-to-halo variance, and halo mass. I will utilize the statistical power of the new DREAMS Cold Dark Matter simulation suite, which is comprised of 1024 zoom-in Milky Way-mass halos with varied initial conditions as well as cosmological and astrophysical parameters. Applying a normalizing flows emulator to these simulations, the uncertainty in the local dark matter speed distribution is dominated by halo-to-halo variance and, to a lesser extent, uncertainty in host halo mass. Uncertainties in supernova and black hole feedback (from the IllustrisTNG model in this case) are negligible in comparison. Using the DREAMS suite, I will present a state-of-the-art prediction for the dark matter speed distribution in the Milky Way. Although the Standard Halo Model is contained within the uncertainty of this prediction, individual galaxies may have distributions that differ from it. Lastly, I will discuss applying the DREAMS results to the XENON1T experiment and demonstrate that the astrophysical uncertainties are comparable to the experimental ones, solidifying previous results in the literature obtained with a smaller sample of simulated Milky Way-mass halos.

Towards Probabilistic Cataloging with BLISS: the Bayesian Light Source Separator

• Camille Avestruz (University of Michigan–Ann Arbor - Department of Physics)

Room: PI/4-405 - Bob Room Stage-IV dark energy wide-field surveys, such as the Vera C. Rubin Observatory Legacy Survey of Space and Time (LSST), will observe an unprecedented number density of galaxies. As a result, the majority of imaged galaxies will visually overlap, a phenomenon known as blending. Blending is expected to be a leading source of systematic error in astronomical measurements. We present the Bayesian Light Source Separator (BLISS), a framework for probabilistic detection, deblending, and measurement. We demonstrate the potential of this method with numerical experiments using synthetic observations where truth is known. We highlight experiments that show (i) robustness to spatially varying backgrounds and point spread functions, (ii) how propagating the probabilistic detections to per-object flux posteriors substantially improves aperture flux residuals, and (iii) how our method retains accurate and well-calibrated posterior approximations for shear estimation under increasingly complex observational systematics. BLISS is a scalable, uncertainty-aware tool for mitigating blending-induced systematics in next-generation cosmological surveys.

An Empirical Probabilistic Model for Field-Level Galaxy Distributions in Tomographic Density Slabs

• Supranta Sarma Boruah (University of Pennsylvania)

Room: PI/4-405 - Bob Room Upcoming photometric surveys such as LSST, Roman, and Euclid will map billions of galaxies, opening the door to field-level cosmological analyses. In this talk, I will present a density-slab framework for performing a field-level analog of the standard 3x2pt analysis, where we model galaxy and weak lensing observables by forward modeling density slabs of O[100 Mpc] at the map level rather than through two-point correlation functions. A central challenge in this program is modeling the field-level distribution of galaxies down to small scales (a few Mpcs), where the galaxy–matter connection is non-linear, non-local, non-Poissonian, and correlated across different tracer populations. I will describe our empirical probabilistic model of galaxy bias that captures all of these effects through a flexible parametric form calibrated against simulations. I will show that the model accurately reproduces the statistical properties of galaxy fields from both the UniverseMachine and IllustrisTNG simulations, demonstrating its robustness across different galaxy formation prescriptions and its viability as a forward model for field-level inference of photometric surveys. I will also briefly discuss a related model for the joint non-Poissonian distribution of multiple galaxy populations within halos in the Halo Occupation Distribution framework.

A Point-Transformed Gaussian Model for Field-Level Mass Distributions in Tomographic Density Slabs

• Alexander Tong (University of Pennsylvania)

Room: PI/4-405 - Bob Room The work presented in this talk is part of a program of performing field-level cosmological analyses using galaxy and weak lensing observables in density slabs of $O[100\ \mathrm{Mpc}]$. The inference pipeline requires a model for the mass overdensity field which is accurate to scales of a few Mpc across cosmologies, and also fast in the generation of sample fields. I will describe a scheme in which the nonlinear mass overdensity field is obtained by applying some point transformation to a Gaussian random field. I will present our transformation function that, with a small number of parameters, characterizes the mass distribution in a slab for different redshifts, slab widths, and cosmologies. I will show that the model accurately reproduces the statistical properties of mass overdensity fields from Gower Street simulations, demonstrating its viability for field-level inference. Finally, I will discuss a validation test which uses the point-transformed Gaussian model to infer the cosmology from the output of an $N$-body simulation.

SPT-3G: Cosmology from CMB Lensing and Delensed EE Power Spectra Using 2019-2020 Polarization Data

• Fei Ge (CalTech)

Room: PI/4-405 - Bob Room I will present the cosmological analysis from the simultaneous Bayesian estimates of gravitational-lensing potential bandpowers and unlensed cosmic microwave background (CMB) EE bandpowers directly using the polarization maps from the South Pole Telescope (SPT) observed in 2019/20. These observations produce the deepest high-angular-resolution CMB polarization maps at 90, 150, and 220 GHz to date, making the standard Quadratic Estimation (QE) method suboptimal for lensing reconstruction. In this analysis, we use the Marginal Unbiased Score Expansion (MUSE) method, which is an optimal map-level Bayesian inference method for CMB lensing potential bandpowers and unlensed CMB EE bandpowers, effectively using all N-point statistics of the CMB polarization maps. The constraints on the Hubble constant (H0) and the amplitude of structure growth (S8) from this work are comparable to those from Planck using full-sky temperature and polarization observations, enabling a powerful test of the LCDM model. With the lensing potential bandpowers reconstructed from the CMB polarization signal, we test the anomaly of excess lensing power from the LCDM prediction, and detect the impact of non-linear structure evolution on CMB lensing. We also explore the extensions of the LCDM models.

Robust CMB B-mode analysis with Needlet-ILC and simulation-based inference

• Adriaan Duivenvoorden

Room: PI/4-405 - Bob Room Polarized Galactic emission is the foremost challenge for searches for a background of primordial gravitational waves imprinted in the polarization of the CMB. We argue that current methods struggle to address this challenge, either by being overly susceptible to model misspecification, or by failing to properly propagate the uncertainty due to residual Galactic emission after foreground cleaning. To address these issues, we explore a novel analysis framework for parameter inference with large-scale CMB polarization data. Our method combines simulation-based inference with the needlet internal linear combination (NILC) algorithm to compress the data into a summary statistic that is robust to model misspecification and small enough for neural posterior estimation with normalizing flows. We show that the semi-blind nature of the NILC-based compression significantly increases robustness to mismodeling of the anisotropic and non-Gaussian properties of the foreground fields. Using an idealized ground-based setup inspired by the Simons Observatory Small Aperture Telescopes, we demonstrate improved statistical constraining power for the tensor-to-scalar ratio r and improved robustness to complex foregrounds compared to other techniques in the literature. Trained on a semi-analytical foreground model, the method yields unbiased results across a range of PySM simulations, including the high-complexity d12 model, for which we obtain r=(1.09±0.27)⋅1e−2 for input r=0.01 and sky fraction fsky=0.21. Our results highlight the importance of designing data compression schemes for SBI that prioritize robustness to model misspecification over statistical optimality, and demonstrate the feasibility and advantages of a complete maps-to-parameters simulation-based analysis of large-scale CMB polarization for current ground-based observatories.

Modeling high-redshift tracers at the field level

• Roger de Belsunce (MIT)

Room: PI/4-405 - Bob Room Current and future high-redshift spectroscopic surveys such as DESI, DESI-II, and Spec-S5 raise the question of how to fully extract the information contained in these datasets. Field-level inference is opening a new frontier in cosmology by enabling analyses that capture the full information content of cosmological observations, rather than relying on two- or three-point summary statistics. High-redshift surveys probe unprecedented cosmological volumes, access quasi-linear modes sensitive to fundamental physics, benefit from precise calibration against small-scale hydrodynamic simulations, and exhibit diminishing shot noise toward higher redshift. However, accurately modeling these observations — particularly at the field level — remains a major challenge. I will present recent progress in modeling the Lyman-alpha forest at the field level and compare the information content of joint power spectrum and compressed bispectrum analyses with field-level inference approaches, combining methods from effective field theory (EFT) and machine learning (ML). This framework is then extended to additional high-redshift tracers, including Lyman-break galaxies and Lyman-alpha emitters, whose cross-correlations unlock new opportunities for extracting cosmological information and controlling systematic uncertainties. Finally, I will present ongoing work that combines perturbative mock generation with generative ML techniques to bridge large and small scales, enabling simulations that span DESI-like cosmological volumes while retaining the small-scale structure captured by hydrodynamic simulations. Together, these developments provide a scalable path toward fully exploiting next-generation surveys such as DESI-II and Spec-S5, establishing high-redshift structure as a key arena for precision, field-level cosmology.

High-redshift galaxies at the field level

• James Sullivan (MIT)

Room: PI/4-405 - Bob Room The future of large-scale structure is at high redshift. In particular, the surveys of the next decade that will be at the vanguard of precision cosmology will target star-forming galaxies at z>2. Field-level inference efforts will, therefore, inevitably turn toward high-redshift samples in the near future, the clustering of which is not yet well-understood. The computational intractability of hydrodynamical simulations means that EFT-bias models at the field level will be especially compelling choices of field-level inference forward models for these samples. I will present my recent work measuring the EFT-bias parameters of simulated high-redshift galaxies at the field-level for the first time. I will also discuss how these results can be leveraged toward building simulation-based priors, which aid current field-level inference efforts.

Simulation-based inference from the Lyman-alpha forest 1D power spectrum with CAMELS

• Francesco Sinigaglia (Institute for Fundamental Physics of the Universe / SISSA)

Room: PI/4-405 - Bob Room We perform for the first time full simulation-based inference on the Lyman-$\alpha$ forest 1D power spectrum. In particular, we consider the prediction of the Lyman-$\alpha$ forest $P_{\rm 1D}(k)$ at $2.0<z<3.5$ from the \texttt{CAMELS} cosmological hydrodynamic simulations run with the \texttt{IllustrisTNG} and \texttt{SIMBA} galaxy formation models. We train a normalizing flow to perform neural posterior estimation of two cosmological parameters ($\Omega_m$ and $\sigma_8$) and four astrophysical parameters parametrizing supernova and AGN feedback. When training and testing the neural network on the same baryon physics model, the posterior distributions of the cosmological parameters are found to be in excellent agreement with the true parameters values (within $10\%$ deviations in $\gtrsim 75\%$ and $\gtrsim 90\%$ of the cases for $\Omega_m$ and $\sigma_8$, and a precision better than $10\%$ in both), while the astrophysical parameters are generally unconstrained due to the limited probed volume. When training on one model and testing on the other (e.g., training on \texttt{IllustrisTNG} and testing on \texttt{SIMBA}, or viceversa), the performance is significantly worse, both in accuracy and in precision, resulting in a $\sim 10\%$ positive bias on the predicted values for $\sigma_8$. We show that a multi-domain training based on the combination of simulations from both models recovers unbiased constraints, offering an effective solution to cope with the complex problem of the lack of convergence in the predictions from different galaxy formation models. This study represents a promising way forward to constrain cosmology and fundamental physics with the Lyman-$\alpha$ forest with artificial intelligence. In addition, we showcase the ongoing application of the same framework to the DREAMS simulations, to jointly constrain astrophysics, cosmology, and the nature of dark matter.

Cosmological Inference from Galaxy Populations with Machine Learning

• Natali de Santi

Room: PI/4-405 - Bob Room Galaxies are the primary tracers of the large-scale structure of the Universe and are traditionally used through summary statistics such as correlation functions and power spectra to constrain cosmological models. However, galaxy populations themselves contain rich information through their spatial distribution, environments, and internal properties, potentially extending beyond the information captured by standard summary statistics. In this talk, I will present a series of machine learning approaches designed to extract cosmological information directly from galaxy catalogs. I will discuss work using graph neural networks (GNNs) to infer cosmological parameters from simulated galaxy populations, including studies demonstrating robustness to observational effects and domain shifts across semi-analytic and hydrodynamical galaxy formation models. I will also present recent work exploring the cosmological information content of individual galaxies using symbolic regression, as well as ongoing efforts based on probabilistic generative models. Together, these results suggest that galaxy populations may encode cosmological information in ways that complement traditional large-scale structure analyses, opening new avenues for field-level and simulation-based inference in upcoming surveys

Discussion: Robust forward modelling for real data and beyond Λ-CDM Room: PI/4-405 - Bob Room

CLOSING REMARKS Room: PI/4-405 - Bob Room