Research

My publications can be found here on iNSPIRE HEP.

My Projects

• Reconstruction of Continuous Cosmological Fields from Discrete Tracers with Graph Neural Networks (paper): In collaboration with Yurii Kvasiuk, Prof. Matthew Johnson, and Dr. Moritz Munchmeyer, we trained a machine learning algorithm on galaxy properties to reconstruct 3D cosmological matter fields. We used a hybrid Graph Neural Network – Convolutional Neural Network (GNN-CNN) architecture and obtained an accurate prediction of the dark matter and electron density fields. This provides a method to use luminous (observable) discrete tracers like galaxies to infer the unobservable continuous fields.

• Reconstruction of Dark Matter and Baryon Density From Galaxies: A Comparison of Linear, Halo Model and Machine Learning-Based Methods (paper): This is a follow-up to the above paper, where we compared different methods to reconstruct dark matter and baryons from galaxy data. The traditional methods we considered are a linear transfer function, NFW painting, and mass-weighted galaxy grid assignment. For machine learning approaches, we used our GNN-CNN model and made a GNN NFW painting model. We found that the GNN-CNN method is superior to halo-model based methods on all scales. Additionally, we proposed a technique that would allow the GNN-CNN model to be used for parameter inference, most notably to assess whether astrophysics feedback is small or large compared to the training dataset.

• Constraints on Cosmology Beyond LCDM with Kinetic Sunyaev Zel’dovich Velocity Reconstruction (paper): The kinetic Sunyaev Zel’dovich (kSZ) effect is a CMB secondary anisotropy contribution caused by Thomson scattering from the bulk-motion of free-electrons. In kSZ tomography, or kSZ velocity reconstruction, the remotely observed CMB dipole is reconstructed by using the statistically anisotropic cross-correlation of kSZ with a galaxy survey. In a project with Prof. Matthew Johnson and Dr. Selim Hotinli, we considered what cosmological information we can obtain from a kSZ velocity reconstruction analysis using Planck and unWISE data. We placed constraints on local-type primordial non-Gaussianity and baryon-dark matter isocurvature, which illustrates that with upcoming data kSZ velocity reconstruction will be a powerful probe of these models. As well, we obtained the most powerful constraints to date on the intrinsic CMB dipole, super-horizon matter-radiation isocurvature, and void models. The void constraints are also more broadly representative of constraints on large-scale inhomogeneities such as cosmic bubble collisions. The beyond-LCDM models that we constrained are a small sample of the potential fundamental physics that can be done with upcoming CMB secondaries.

• Improving Photometric Galaxy Clustering Constraints with Cross-Bin Correlations (paper): With Dr. Jessie Muir and Prof. Matthew Johnson, we considered the modeling pipeline of existing and future galaxy surveys to identify potential gains in constraining power for galaxy clustering constraints. The key change we considered was the impact of incorporating cross-bin correlations into the clustering analysis for DES Y3 and LSST Y1, using Fisher forecasting techniques. Moreover, we investigated how accounting for RSD and magnification could increase the benefit of using cross-bin correlations and if the systematic biases from photometric redshift uncertainties outweighed this gain.

• Cosmological Measurements from the CMB and BAO are Insensitive to the Tail Probability in the Assumed Likelihood (paper): In collaboration with Prof. Will Percival, Dr. Simone Paradiso, Dr. Alex Krolewski, and Dr. Shahab Joudaki, I investigated the robustness of cosmological constraints to statistical assumptions made about the tail probability of the likelihood for the Cosmic Microwave Background (CMB) and Baryon Acoustic Oscillations (BAO). Typically, a Bayesian analysis is used when fitting cosmological models to data which requires assuming a prior for the model and the form of the likelihood. We assessed the sensitivity of cosmological constraints to changes in the likelihood assumptions. First, for the case of a covariance matrix drawn from a finite number of mock catalogs which adds uncertainties that replace the common Gaussian likelihood with a multivariate t-distribution. We then considered a more general sensitivity test to the typical assumption of Gaussian likelihoods by determining the impact of increasing the size of the likelihood tails by using different t-distributions.

• Stochastic Gravitational Wave Background: I worked under the supervision of Dr. Ghazal Geshnizjani during a summer undergraduate research internship at the Perimeter Institute for Theoretical Physics in 2020. In this project, we examined the cross-correlation between a theoretically predicted stochastic gravitational wave background and galaxy clustering.