The vast majority (around 85%) of the matter in the Universe is invisible. The search for that invisible matter, so-called dark matter, is one of the most pressing open questions in all of physics. Noble liquid time-projection chambers are a leading direct-detection technology to search for dark matter. The Noble Element Simulation Technique (NEST) is a premier Monte Carlo software for modeling the microphysics of these detectors. My research aims to improve NEST microphysics in the low-energy regime for the electronic recoil (ER) model, critical for dark matter searches. In this talk, I will present the details of my optimization approach to NEST and showcase results that provide a novel, unified field-dependent parameterization of the NEST ER model, continuous with the high-energy regime. This will directly strengthen background noise discrimination and sensitivity for current and next-generation dark matter experiments.

As machine learning systems are increasingly used to make consequential decisions in areas like hiring, lending, and criminal justice, ensuring these systems treat different groups equally has become a critical challenge. Existing approaches to build fair AI models either cannot guarantee fairness will be satisfied, or sacrifice the ability the effectively learn from the data. This project introduces the "fairness layer", a new neural network component that mathematically guarantees a choses notion of fairness is satisfied whenever the model makes predictions, while still allowing the model to learn end-to-end from data (unlike some previous methods). We also introduce an algorithm designed for real-world deployment scenarios where predictions must be made in small, rapid batches, ensuring fairness not only holds batch-by-batch, but also cumulatively over time. Experiments across loan default prediction, employee wage modeling, and synthetic datasets show that the fairness layer consistently outperforms existing approaches, achieving both lower error and guaranteed fairness compliance. This work offers a practical tool for deploying accurate and fair AI systems that are not only accurate but verifiably fair, with implications for any domain where algorithmic designs affect people's lives.

The nucleus of a heavy atom can exhibit discrete energy states analogous to electronic shells; nuclear states typically have excitation energies in the keV to MeV range. Th-229 uniquely possesses an exceptionally low-lying nuclear isomeric state, approximately 8 eV above the ground state, making it excitable by VUV lasers. One of our projects involves studying Th-229 within thorium dioxide (ThO2) thin films electrodeposited on stainless steel under ultra-high vacuum conditions. Nuclear excitation is measured via internal conversion, whereby excited nuclei transfer their energy to valence band electrons and thus eject them as detectable free electrons collected by micro-channel plates (MCPs). A central motivation for this work is that Th-229 with its high transition frequency and long lifetime is critical for developing a nuclear clock, which is potentially far more stable than current atomic clocks. The quality factor Q, which is the ratio between central transition frequency and the linewidth of the transition (inversely proportional to life time), indicates how well the clock can define time. The Th-229 has a transition frequency of over 2000 THz and 12 microsecond lifetime resulting in a relatively high Q. Such potential is only to be fully realized given a high signal to noise ratio. Noise in the frequency measurement propagates through to clock performance, and will limit interrogation fidelity. This constrains its applicability in searching for new physics like testing temporal variation of fundamental constants.

My research with Dr. Justin Baker is in the area of dynamical systems. Our work focuses on data-driven methods for these systems in partially observed settings. These include HAVOK (Hankel Alternative View Of Koopman) and projection operator methods based on the Mori-Zwanzig formalism. Our goal is to unify data-driven time series methods with the rigorous principles of physics based models. In particular, we have worked to formalize the relationship between regression and Mori-Zwanzig projection operators, examining how to guarantee theoretical principles in applied settings. This work has implications for mathematical modeling generally with applications to diverse fields including climate modeling, physics simulations, and artificial intelligence.