Research 

Student Team Members: Alexander Kuhn, Bonnabelle Zabelle 

Project Faculty Mentors: Prof. Claudia Scarlata, Prof. Sara Algeri, Prof. Galin Jones. 

Project Summary: The problem of detecting spectral lines is ubiquitous in several areas of astronomy, as they enable measurements of fundamental physical properties such as distances, chemical compositions, temperature and more. In the near future, the Euclid satellite will provide spectra for millions of galaxies for which distances need to be measured from emission lines. While several approaches have been proposed in the astronomy literature to address this issue, from a statistical point of view, they all lack the statistical rigor necessary to control for the inflation of the probability of false discovery occurring when the search is conducted by scanning multiple regions of the spectrum considered – a statistical phenomenon also referred to in high-energy physics as ``the look-elsewhere effect’’. Motivated by the above limitations, this project aims to extend existing (frequentist) solutions proposed to handle the look-elsewhere to enable both the detection and the identification of multiple spectral lines. Moreover, the students will develop a brand-new Bayesian solution to address the problem.The validity and usefulness of the statistical advancements entailed by this project, as well as adequate comparisons of frequentist and Bayesian proposed procedure, will be demonstrated through simulation studies, and will be applied to data generated from the new Euclid  and  James Webb Space telescopes. 

Student Team Members: Jacynda Alatoma and Elliott Tanner 

Project Faculty Mentors: Priscilla Cushman 

Project Summary: The SuperCDMS experiment employs modular cryogenic solid-state detectors to search for dark matter. The project uses Machine Learning (ML) techniques to address challenges in modeling pulse shapes resulting from particle interactions. As part of the FAIR4HEP initiative, this study leverages high-energy physics to develop community-wide FAIR frameworks for AI. The accessible dataset encourages further research, and a second dataset with enhanced coverage is in preparation to explore ML techniques' limitations on more complex data. 

Student Team Members: Abby Stokes, Lexi Leali, Zhizhong Guan. In addition to Haowen Zhong (DSMMA Cohort 2022) and Daniel Kukla (undergraduate student) 

Project Faculty Mentors: Vuk Mandic and Michael Steinbach 

Project Summary: Parameter estimation in the study of stochastic gravitational waves background is a crucial task. Traditional Bayesian analysis relies on a likelihood function, determined by our understanding of the underlying physics process. In directional search, the likelihood function lacks a closed form which furthermore makes the parameter estimation quite complicated. Leveraging Machine Learning, particularly Simulation Based Inference and Normalizing Flow, enables direct estimation of posterior distributions of parameters from observation data, bypassing the explicit need for a likelihood function. In our project, we will focus on isotropic search and directional search separately and try to use NF to recover the injected parameters from observations. 

Student Team Members: Matthew Choquette, Marianne Peterson

Project Faculty Mentors: Lindsay Glesener, Jarvis Haupt

Project Summary: Solar flares are known to release energy and accelerate particles through mechanisms which are not fully understood. The Focusing Optics X-ray Solar Imager (FOXSI) is a sounding rocket payload that will attempt to observe a solar flare in X-rays after a trigger has identified that a flare has begun. The trigger is based on several predictors, including solar X-ray flux and temperature. This project aims to use machine learning methods to enhance the existing trigger’s accuracy and explore the use of UV and microwave data as additional flare predictors. 

Student Team Members: Brendan King, Angelo Rustichini, and Taarak Shah 

Project Faculty Mentors: Vuk Mandic, Claudia Scarlata, and Sara Algeri 

Project Summary: This project aims to correlate sky maps of different electromagnetic signals, for example from the cosmic microwave background (CMB) or gravitational lensing (GL), with the maps of energy density in the stochastic gravitational wave background (SGWB). If such correlations are observed, they would provide information about the formation and evolution of structure in the universe and potentially provide a way of distinguishing different contributions to the SGWB. The project will develop a statistical formalism to measure the angular power spectrum of the cross-correlations.

Student Team Members: Alexander Criswell and Jesse Miller

Project Faculty Mentors: Vuk Mandic and Galin Jones

Project Summary: This project aims to further our understanding of the nature of matter at the heart of neutron stars by combining an ensemble analysis of binary neutron star post-merger gravitational wave emission with inspiral gravitational wave data, electromagnetic observations, and empirical relations derived from numerical relativity simulations. To accomplish this, the team will: 1. Synthesize hierarchical Bayesian methods with existing single-event post-merger analyses. 2. Inform the analysis with a multimessenger approach, using electromagnetic and gravitational wave data alongside results of numerical relativity simulations. 3. Validate this new method by recovering simulated signals and apply it to real data.

Student Team Members: Laura Salo, Rui Zhou, Sam Johnson

Project Faculty Mentors: Pat Kelly and Galin Jones

Project Summary: The Laser Interferometer Gravitational-Wave Observatory (LIGO) has now detected approximately fifty mergers of binary systems of black holes and neutron stars. Explaining the population of binary systems revealed by LIGO will require an improved understanding of the evolution of interacting massive stellar systems. A primary goal of the proposed research is to understand the reason for this pattern, and its implications for the creation of massive binary systems whose remnants are observed by LIGO. By modeling the emission lines of nearby galaxies that have hosted multiple SNe, the team hopes to characterize the massive stellar populations of the galaxies that have hosted multiple stripped-enveloped SNe, and multiple examples of hydrogen-rich SNe. 

Student Team Members: Becca Dura, Sai Sharan Sundar, Shaowei Wu

Project Faculty Mentors: Dr. Greg Pawloski, Dr. Michael Steinbach

Project Summary: The NOvA experiment is designed to answer the fundamental questions on properties of neutrinos. The goal of our project is to develop an algorithm to search for neutrinos in the time frame of gravitational wave or supernova events. We hope that the algorithm will also identify the number of individual neutrino interactions and the individual cluster of hits associated with each neutrino interaction. Our initial idea for this project is to process the full data recorded during a supernova or gravitational wave event in sections, meaning we would only process a small portion of the entire time window at a time and look for correlations between each portion.  

Student Team Members: Katie Gerot, Alex Granados

Project Faculty Mentors: Vuk Mandic, Sara Algeri, Galin Jones

Project Summary: To date, only upper limits for the anisotropic stochastic gravitational wave background (SGWB) have been reported. Different electromagnetic (EM) tracers (such as galaxy counts or weak lensing) have the potential to aid in parsing through the noise of current data from LIGO. Throughout this project we aim to use cross-correlation analysis of SGWB with EM tracers to indirectly detect the SGWB and probe different anisotropic generators.

Student Team Members: Kiet Pham and Sam Penders

Project Faculty Mentors: Professor Vuk Mandic and Dr. Michael Steinbach 

Project Summary: The long-lived (> 10s) and narrow-band gravitational wave (GW) signals regime has been studied in the context of a variety of astrophysical processes, most notably, in newborn star and black hole accretion disks following stellar collapse. These more exotic sources cannot be accurately modeled, owing to theoretical uncertainties. Such signals can be detected by clustering excess GW power in the time-frequency spectrograms (tf-maps) of cross-correlated data. This project aims to develop a deep learning pipeline to identify such transient signals, targeting low signal to noise ratio (SNR) ones that are heavily burdened in noise in tf-maps. The technique would be applicable to different search time-scales, especially the days-to-weeks long periods, in which the computational cost is high and a variety of waveforms could potentially be missed by the current detection method.

Student Team Members: Haowen Zhong, Aritra Banerjee, Lauren Wills(Undergraduate) 

Project Faculty Mentors: Vuk Mandic, Xiaotong Shen 

Project Summary: The improved sensitivity of 3G gravitational wave detectors opens the possibility of detecting the primordial cosmological stochastic gravitational wave background(CSGWB). However, the CSGWB will be masked by the foreground generated by a huge amount of compact binary coalescence(CBC). In the previous study, we proved the efficacy of the notching algorithm which is based on the parameter estimation(PE) of every individual CBC event. However, the technique/algorithm that can simultaneously implement PE for ~1 million overlapping CBCs is not available now. Our aim of this project is to directly dig out CBC tracks from given noisy spectrograms by Machine Learning algorithms instead of doing PE.

Student Team Members: Tyler Barna, Ana Uribe

Project Faculty Mentors: Michael Coughlin, Jie Ding 

Project Summary: Kilonovae are the optical counterpart to gravitational waves (GW). The ability to observe a kilonova corresponding to a GW would enable multi-messenger astrophysics. However, the higher detection rate and relative similarity of supernovae (and other potential astrophysical objects) make detecting kilonovae problematic. Identification is made even more difficult by the transient nature of these objects and a finite observing budget on a nightly basis. Currently, these multiple factors are evaluated by individual researchers, introducing potential error while also consuming time that could otherwise be used for higher-level analysis. Being able to quickly and automatically flag kilonova candidates for follow-up observation therefore represents a valuable tool. Our project aims to determine how to choose objects to follow up on, and what objects would most benefit from follow-up observations. We plan to define an algorithm using reinforcement learning to automate these decisions and apply astrophysical concepts in evaluating model parameters.

Student Team Members: Lindsey Gordon, Kat Kompanets

Project Faculty Mentor(s): Tom Jones (Astrophysics), Michael Steinbach (Data Science) 

Project Summary: WOMBAT is an MHD simulation suite used to study astrophysical fluid motions and interactions, which produces a wide range of radiation types, from radio waves to cosmic rays. It produces very large 3D outputs of various state variables (density, pressure, electromagnetic fields, etc.). Fluid interactions produce a number of interesting features and structures, such as shocks, rarefactions, shear, and filaments. These structures are typically identifiable by eye, but are not necessarily simple to identify algorithmically, especially in three dimensions and/or at oblique angles. We are developing a supervised classifier for WOMBAT outputs to locate and classify fluid features to speed up our simulation analyses. This work will also act as a foundational precursor to a long-term goal of the new version of the simulation (wombatwisdom), which hopes to incorporate in-situ machine learning classification to identify structures being produced as the simulation runs. A major application of this is to save computational resources by identifying runs that are behaving aphysically and terminating them early.