Somayeh Khakpash

I am a Visiting Assistant Professor of Physics at Lehigh University with extensive experience in astronomical data analysis and machine learning. I am the coordinator of the Rubin Observatory Transient and Variable Stars Science Collaboration (TVSSC) Microlensing Subgroup. The Rubin Observatory will shape how we explore the universe by delivering the deepest, widest, and fastest optical survey ever conducted, enabling unprecedented discoveries in fields ranging from dark matter and dark energy to transient phenomena like supernovae and microlensing.
As a former LSSTC Catalyst Fellow at Rutgers University, I developed a deep learning autoencoder for magnification maps of lensed quasars and supernovae. This model significantly improves the efficiency and speed of analyzing lensed quasar light curves by enabling fast generation of magnification maps — a critical capability for the Rubin Observatory.
Prior to Rutgers, I was a postdoctoral researcher at the University of Delaware, where I created data-driven templates for stripped-envelope supernovae to classify stellar explosions photometrically and identify unusual behavior. I completed my PhD and MS in physics at Lehigh University, where I worked on developing autmated algorithms to extract features from microlensing light curves and enhance their detection, classification, and charachterization.
My research combines algorithm development, feature extraction, and statistical modeling to detect, classify, and characterize time-dependent astronomical events efficiently. As subgroup coordinator, I coordinate efforts to enhance simulations, develop software for detection and classification of microlesing events with Rubing, and explore synergies with other surveys like the Roman Space Telescope.
I am also a member of the Roman Galactic Exoplanet Survey Project Infrastructure Team, preparing for the Galactic survey of NASA’s Roman Space Telescope. In addition, as a former LSSTC Data Science Fellow, I gained expertise in data engineering, Bayesian statistics, machine learning, image processing, and advanced data visualization.

My research addresses fundamental astrophysical questions about the hidden populations of planets, black holes, and the dynamic universe, focusing on two upcoming major astronomical surveys: the Roman Space Telescope and the Vera C. Rubin Observatory’s Legacy Survey of Space and Time (LSST). These projects will completely change our understanding of the cosmos by capturing vast amounts of data over time, allowing us to discover and study transient astronomical events like galactic and extra-galactic microlensing, supernovae (SNe), and other dynamic phenomena. Microlensing is a unique technique that helps uncover hidden objects in space, such as free-floating planets and black holes, by observing how their gravity bends and magnifies the light from distant stars. This method is essential for discovering planetary systems that would otherwise be impossible to detect. With the Roman and Rubin surveys, which will monitor millions of stars over long periods, my work aims to develop automated tools that will allow us to efficiently detect and analyze these rare events. I began my research career focusing on planetary microlensing, where I developed automated, efficient methods to detect and characterize microlensing events in massive datasets. This work enabled faster and more accurate analysis of planetary systems, advancing our understanding of planets beyond our solar system. I have since expanded my research to include supernovae, where I have developed techniques to understand and interpret these stellar explosions using data from many different surveys over time. As a graduate of the LSST Discovery Alliance Data Science Fellowship Program, I mastered approaches aimed at storing, analyzing, and visualizing large volumes of data, including time series and images. I became an expert in Machine Learning, statistics, and Data Science. Combined with my extensive experience working on light curves (time series of brightness of astronomical objects over time) of various astronomical events, applying statistical and machine-learning methods, I have created novel models for the detection, characterization, and interpretation of data from large photometric surveys. Furthermore, as an LSST Discovery Alliance Catalyst Postdoctoral Fellow, I have gained extensive experience with deep learning models applied on image datasets. Here are some of my main publications: