Dissertations, Theses, and Capstone Projects

Date of Degree

6-2026

Document Type

Master's Thesis

Degree Name

Master of Science

Program

Astrophysics

Advisor

Shy Genel

Advisor

Francisco Villaescusa-Navarro

Subject Categories

Artificial Intelligence and Robotics | Astrophysics and Astronomy | Cosmology, Relativity, and Gravity | Data Science | External Galaxies

Keywords

cosmological parameter inference, galaxy kinematics, large-scale structure, machine learning, statistical methods, velocity fields

Abstract

We perform field-level likelihood-free inference of the matter density parameter Ωm from simulated galaxy catalogs using machine learning models with differing inductive biases. Using features extracted from hydrodynamic simulations in the CAMELS suite, we investigate how both observable choice and model architecture govern the extraction of cosmological information. We consider galaxy positions and line-of-sight peculiar velocities, both separately and in combination, and compare permutation-invariant Deep Sets, implemented with either standard multilayer perceptrons (MLPs) or Kolmogorov–Arnold Networks (KANs), to graph neural networks (GNNs) implemented with MLPs, which explicitly encode spatial relations. We evaluate inference performance under both in-distribution and out-of-distribution (OOD) conditions by testing across simulations with different subgrid galaxy-formation prescriptions. We find that Deep Sets can infer Ωm from in-distribution data with a mean relative error of approximately 18% and from OOD data with a mean relative error of ∼25% using only line-of-sight peculiar velocities, with KANs and MLPs achieving comparable performance. This indicates that peculiar velocities encode nearly robust cosmological information accessible to permutation-invariant models. When positional information is included, Deep Set performance remains unchanged, while GNNs outperform them, inferring Ωm with a mean relative error of about 10% in-distribution and 10%–17% OOD. These results clarify how observable choice and architectural inductive bias jointly determine the robustness and effectiveness of machine-learning-based cosmological inference, with direct implications for the design of analysis pipelines for upcoming galaxy surveys.

Download

Share

COinS