Physical Sciences and Engineering

Optimal Thermal Management Architectures for Electrified Propulsion Aircraft

Maxfield Arnson, Aerospace Engineering

Electrified aircraft have the potential to reduce the aviation sector’s impact on the Earth’s climate. A unique challenge when designing these aircraft is the management of the waste heat produced by electric machines such as motors, power electronics, and batteries. Conventional aircraft dispose of any waste heat generated onboard the vehicle by expelling it from the engines alongside the exhaust gases. Because electrified aircraft decouple their power production from their thrust production, they require dedicated thermal management systems. The thermal management and power production subsystems are highly coupled, and the optimal design of these depends on their arrangement, or “architecture.” Maxfield’s thesis develops a methodology to couple the thermal management and power production systems, in addition to methods which isolate the architectures which best reduce vehicle-level energy consumption.

Engineering Nano-Bio Interfaces for Functional Materials and Devices

Mohammad Asadi Tokmedash, Chemical Engineering

Implant-associated complications, such as infection, adverse immune responses, and poor tissue integration, remain major causes of failure, while chemical-based strategies offer limited durability and risk toxicity or resistance. This dissertation develops a scalable, tunable biomaterial platform that uses engineered surface topography to address these challenges through physical cues alone. Employing bottom-up layer-by-layer assembly combined with mechanical nanomanufacturing, I generate nano- and micro-scale topographies with independently tunable size, geometry, and mechanical compliance, decoupling physical structure from surface chemistry. These topographies establish generalizable structure-function design rules that suppress bacterial adhesion while promoting regenerative host responses. Integrated in vitro and in vivo murine implantation studies demonstrate long-term antibacterial activity, immune modulation toward pro-healing phenotypes, and enhanced bone regeneration. This drug-free approach provides a broadly applicable, translatable platform for designing multifunctional implant surfaces that reduce infection, improve tissue integration, and extend device longevity.

Multicomponent Diffusion and Diffusive Isotope Fractionation in Natural Silicate Melts

Bobo Bai, Earth and Environmental Sciences

The differential movement of chemical elements within molten rock controls volcanic eruptions, magma evolution, and crust formation. This dissertation investigates how major elements and their isotopes diffuse in natural silicate melts using piston-cylinder experiments and numerical modeling. Chapter 1 introduces the basic concept of multicomponent diffusion and diffusive isotope fractionation. Chapter 2 tests whether multicomponent diffusion in basalt can be described using a single, temperature-independent model. Chapter 3 extends this investigation to andesite, develops methods to account for non-uniform starting compositions in diffusion experiments, and evaluates whether diffusion eigen-components are composition-independent from basalt to andesite. Chapter 4 explores diffusion in dacite to assess the broader applicability of a universal eigen-component framework. Chapter 5 investigates diffusive titanium isotope fractionation in natural andesitic melt. Together, this work provides a simpler and more predictive model of chemical transport in natural silicate melts.

Toward Globally Inclusive and Safe AI Systems Through Cultural and Behavioral Alignment

Angana Borah, Computer Science and Engineering

Large Language Models (LLMs) are increasingly used by people to seek information and interact online; however, they are primarily built on data representing WEIRD (Western, Educated, Industrialized, Rich, and Democratic) populations. This leads to systems that may reinforce stereotypes, misrepresent cultural norms, and amplify misinformative content. My dissertation develops computational frameworks to evaluate and improve LLM behaviors so they support global inclusivity, truthfulness, and social well-being. I approach this through three complementary research directions: (i) developing cross-cultural awareness: to understand how models encode societal biases and how diverse cultural perspectives in multi-agent modeling can improve outputs; (ii) investigating LLM behavior in social interactions: analyzing group-based bias amplification, polarization, and persuasive dynamics, with curiosity as a positive moderator and (iii) designing lightweight, scalable mitigation strategies: that can be layered onto existing models to improve safety without costly retraining. This work advances practical approaches toward AI systems that empower society.

Enabling Real-World Assistive Agents: From Live Vision to Proactive Context-Aware Information Delivery

Ruei-Che Chang, Computer Science and Engineering

Interacting with the real world is a fundamental part of daily life, yet it remains challenging for individuals who are blind or visually impaired (BVI). It demands live, contextual understanding of dynamic environments, along with communication to fulfill their needs. My dissertation develops assistive AI systems that observe surroundings through multimodal sensing, reason about essential information, and deliver human-like communication. First, I explored design insights through interactions between BVI users and video AI systems in real-world settings. Second, I developed a mobile application that analyzes live camera feeds to generate real-time descriptions aligned with user goals, delivering them in harmony with the audio environment. Lastly, I extend it with human-like capabilities, such as memory, spatial understanding, and the ability to infer user intent, to act as a long-term assistive companion. Ultimately, my dissertation advances a paradigm shift from digital to real-world assistive agents that enhance the independence of BVI individuals.

Engineering Synthetic Cell–Living Cell Communication for Stimuli-Responsive Drug Delivery

Samuel Chen, Mechanical Engineering

Synthetic cells, non-living, cell-sized structures, offer a promising middle ground between systemically released small-molecule drugs and sophisticated yet expensive engineered living cells. This dissertation aims to demonstrate the potential of this approach by developing a synthetic cell drug delivery platform to sense cell-impermeable signals from diseased tissues and release protein therapies. Aim 1 engineers sensing modules that respond to mechanical forces and specific cell-surface ligands using the mechanosensitive channel MscL and a ligand-activated tension-gated channel (LATCH) receptor design. Aim 2 develops two modular downstream secretion pathways, an exocytosis-like mechanism and a cell-penetrating peptide-based protein export system. Aim 3 integrates sensing and secretion modules into a coherent stimuli-responsive synthetic cell. A model cancer treatment is then shown through mechanical force or ligand-binding triggered delivery of the tumor-suppressive cytokine IL-24 to MCF-7 breast cancer cells. Taken together, these tools form the groundwork for a generalizable, modular chassis for context-dependent therapeutic delivery.

Indo-Pacific Climate Interactions and Their Regional Impacts Over the Industrial Era

Yunfan Chen, Earth and Environmental Sciences

The tropical oceans play a central role in regulating global climate variability, yet long-term interactions between the Pacific and Indian Oceans remain poorly constrained due to limited observations. This dissertation investigates Indo-Pacific climate interactions during the Industrial Era, with a focus on the western Indian Ocean, the fastest-warming tropical ocean region over the past century. I develop five new coral records from Tanzania, including the longest available record from this data-sparse region, and combine them with existing proxy archives to reconstruct regional sea surface temperature variability using multiple statistical techniques. These reconstructions reveal pronounced centennial-scale warming and evolving Indo-Pacific climate linkages. To diagnose the underlying physical mechanisms, I analyze climate model simulations to evaluate these Indo-Pacific connections. Finally, I assess the long-term climatic impacts of Indian Ocean variability, particularly its influence on Southeast Asian summer monsoon rainfall, with implications for monsoon predictability and future climate projections.

Energy as a First-Class Resource in Machine Learning Systems Design

Jae-Won Chung, Computer Science and Engineering

Improving the energy efficiency of the execution is a fundamental pursuit. Computing is no exception—and the challenge has intensified with the energy consumption of machine learning datacenters has grown to scales comparable to that of entire cities or even countries. Energy has become a bottleneck resource, driving capital and operational cost, complicating energy procurement, and creating grid reliability challenges.

This dissertation presents systems and techniques for modern machine learning designed around energy, from precise measurement and understanding to large-scale optimization across space and time. In doing so, it establishes the time–energy tradeoff frontier as the central abstraction, which allows independent advancements in each layer of the execution stack to compose upwards and improve the time–energy frontier of the whole stack. Together, these elevate energy as a first-class resource that can be measured, reasoned about, and managed across the full stack, serving as the basis for full stack, principled energy optimization.

Counting Supersymmetric States in AdS/CFT: from Giant Gravitons to Black Holes

Evan Deddo, Physics

The Anti-de Sitter/conformal field theory correspondence (AdS/CFT) is a powerful framework that relates two of the main branches of theoretical physics. On one side of the correspondence is quantum field theory (QFT), which describes the smallest constituents of matter. On the other side are gravitational theories which describe the curvature of space and time. In this work we utilize the superconformal index, a mathematical function that enumerates quantum states in certain QFT models, to investigate gravitational phenomena. In Part I we focus on giant gravitons—large membrane-like objects that exist in theoretical models called supergravity. Information about their quantum fluctuations is encoded in the superconformal index, and we study how these properties arise directly in supergravity. In Part II we study black holes from the perspective of the index. We perform extensive numerical computations to search for evidence of black holes surrounded by graviton gas and giant gravitons.

Quantum-Inspired Analog Computing Architectures for Accelerating Combinatorial Optimization Solvers

Evangelos Dikopoulos, Electrical and Computer Engineering

This dissertation explores quantum-inspired analog computing architectures for solving combinatorial optimization problems. By leveraging concepts from quantum computing, such as dynamical systems, continuous-time operation, and massive parallelism, this research aims to overcome the limitations of quantum and traditional digital architectures. Optimization graphs are directly mapped onto coupled oscillator-based analog compute engines, enabling fully asynchronous, massively parallel operation to deliver unprecedented solution times and energy efficiency. The initial project of this dissertation produced the first oscillator-based LDPC decoder, reframing LDPC decoding as an optimization problem and achieving superior energy efficiency over leading digital LDPC decoders [ESSERC ‘24]. Building on this, a next-generation solver was developed for the Boolean Satisfiability (SAT) problem, a foundational NP-complete optimization problem. This SAT engine features a novel relaxation oscillator-based spin-injection architecture, leveraging physics-inspired heuristics to eliminate the need for problem embedding. It achieves remarkable improvements in solution time and energy efficiency over state-of-the-art solvers [ISSCC ‘25].

The Intersection of Metal–Organic Framework Formation with Formamide Chemistry

Julia Donovan, Chemistry

Metal–organic framework (MOF) syntheses are highly sensitive to synthetic conditions, yet the mechanistic origins of these dependencies remain poorly understood. This thesis establishes formamide chemistry, a ubiquitous MOF solvent, as a central factor in MOF formation. Through systematic variation of water content and headspace composition, the hydrolysis and oxidative decomposition of formamide solvents are directly linked to changes in nucleation kinetics with water concentration affecting phase selection, and headspace composition affecting byproduct formation. These studies reveal previously unrecognized roles of solvent-derived chemistry in directing framework assembly. Building on these mechanistic insights, alternative solvent systems are developed to replicate the chemical function of toxic formamide solvents. This dissertation uncovers unconventional reagent and solvent reactivity, proposes new mechanistic pathways for MOF formation, and provides a foundation for more reproducible and sustainable MOF syntheses.

Optimization and Reinforcement Learning Methods for Supply Chains and Integrated Infrastructure System Design and Operations

Juan-Alberto Estrada-Garcia, Industrial and Operations Engineering

This dissertation develops optimization and reinforcement learning methods for designing and operating supply chains and integrated infrastructure systems under uncertainty. For supply chain risk management, we propose a risk-aware multi-agent framework to identify critical locations and allocate additional resources to hedge against disruptive events. For integrated infrastructure systems, we introduce stochastic static and dynamic models that optimize decisions in data-limited environments. New decision-dependent uncertainty formulations and risk-aware frameworks capture how planning and operational choices influence future system reliability. To ensure tractable computation, we design decomposition algorithms that exploit problem structure and parallelization. We further develop reinforcement learning methods that complement exact optimization by providing approximate, adaptive policies for high-dimensional, dynamic settings. Applications include drone-based infrastructure monitoring under uncertain failures and power system control in wildfire-prone regions. Our approaches can advance scalable, interpretable, and robust decision-making tools for complex integrated systems in modern and future societies.

Mechanisms of Plastic Co-deformation in Nano-scale Heterostructures

Arkajit Ghosh, Materials Science and Engineering

Nano-scale eutectic heterostructures represent a promising class of structural materials, yet they are fundamentally limited by the brittle failure of hard reinforcement phases. This dissertation investigates the scientific principles required to achieve plastic co-deformation, i.e., simultaneous plastic flow of both soft and hard phases, that would overcome the inherent strength-ductility tradeoff. By employing rapid solidification, such as additive manufacturing, along with chemical modification, we demonstrate how structural refinement to nano-scale enables activation of non-traditional deformation mechanisms in normally brittle phases. Through advanced microscopy and nano-mechanical testing, we elucidate the atomic-scale mechanisms that suppress strain localization and delay cracking. Moreover, this project, for the first time, introduces temperature-dependent in-situ mechanical characterization of these heterostructures. This research, therefore, establishes a general framework for architecting robust heterogeneous materials with high damage tolerance, essential for advancing next-generation technologies in automotive, aerospace, or renewable energy materials.

Molecular Structures and Interactions at Buried Interfaces Revealed by Sum Frequency Generation Spectroscopy in Situ in Real Time

Fernando Gomez, Chemistry

Molecular interactions at surfaces/interfaces play significant roles in many industrial applications and impact the environment. One of the greatest challenges of industrial applications is creating industry-leading products while reducing the negative impacts on the environment of their products. This thesis research focuses on investigating buried interfaces related to industrial materials and promoting positive impact of these materials on the environment. Sum frequency generation (SFG) spectroscopy was successfully applied in the studies. SFG is sub-monolayer surface/interface sensitive, which allows for the study of buried interfaces at the molecular level in real time and in situ nondestructively. My SFG results elucidated molecular mechanisms of (1) marine antifouling/fouling release activities of various silicone materials, (2) generally applicable deposition approach of 2D-nanomaterials onto hydrophobic substrates, (3) interactions between organic pollutants and microplastics under aquatic conditions, (4) impacts of surface agents on silicone/plastic adhesion, and (5) emulsification performance of surfactants at the oil-water interface.

Ultra-Thin, Low-Reflection Metasurfaces for Sub-Terahertz Automotive Radar and 6G Beamforming

Ehsan Hafeziasl, Electrical and Computer Engineering

The growing demands of autonomous vehicles and emerging 6G-class wireless links require compact sub-terahertz front ends capable of resolving fine spatial details. Conventional passive hardware manipulates electromagnetic waves through material thickness, curvature, and dielectric properties, resulting in bulky and reflective components that degrade dynamic range and constrain beamforming functions. This dissertation reports the development of ultra-thin metasurface lenses and beamforming devices engineered with periodic arrays of resonant elements. First, a multi-layer 230-GHz metasurface lens is designed, fabricated, and integrated into a radar system, achieving focusing comparable to conventional optics while reducing reflections by an order of magnitude. Second, a reconfigurable, moiré-based metasurface aperture is demonstrated, generating eight orbital angular-momentum modes from a single device with mode purity exceeding 30 dB, thus eliminating the need for multiple fixed apertures. Finally, dispersion-engineered metasurfaces are introduced that link beam direction to frequency, enabling passive beam steering without reliance on phased arrays or moving parts.

Characterization and Modeling of Ductile Fracture and Ultra-Low Cycle Fatigue in Structural Steel at Low Temperatures

Min-Chun Han (Barbour Scholar), Civil Engineering

Ductile fracture and ultra-low cycle fatigue (ULCF) share a common mechanism involving microvoid nucleation, growth, and coalescence under large plastic strains that challenge traditional fracture mechanics. The Void Growth Model (VGM) links microscale void evolution to mesoscopic crack development through a characteristic length scale and has been extended to ULCF as the Cyclic Void Growth Model (CVGM) for simulating structural failure under seismic loading. This dissertation generates low-temperature experimental data without the need for an environmental chamber and provides new insights into ductile fracture and ULCF. It advances SEM-based fractographic characterization methods and develops physically informed fracture models that incorporate temperature effects, improving the assessment of seismic performance in cold regions.

Uncertainty Quantification for Machine Learning via Selective Inference

Yiling Huang, Statistics

Scientific progress depends on replicable results, yet many published findings fail to replicate. A key reason is data-driven decision-making in modern statistical analysis, where researchers select variables or models after querying data and then report standard confidence intervals or p-values. Classical statistical methods assume inferential targets are fixed in advance, and applying them after selection leads to overly optimistic conclusions. Selective inference addresses this challenge by providing valid inference accounting for data-adaptive question formation.

This dissertation develops selective inference methods for a range of modern machine learning procedures. First, we study inference after automated variable selection in regression models, proposing a unified framework for continuous, binary, and count outcomes, and for individual and group selection. Second, we develop inference methods for Gaussian graphical models, enabling valid inference for learned dependencies between variables. Third, we develop hypothesis testing procedures for decision trees, explicitly accounting for the data-adaptive training process.

Multifunctional Adaptive Materials for Robotic Embodied Intelligence

Chuqi Huang (Barbour Scholar), Materials Science and Engineering

Robots operating in confined and dynamic environments face fundamental challenges due to external control and rigid architectures, limiting miniaturization, system integration, robustness, and autonomy. In contrast, biological organisms exhibit “embodied intelligence,” in which fundamental functions (power, actuation, control, etc.) are intrinsically integrated within material structure. Translating this principle to robotics requires novel multifunctional materials, that perform and adapt robotic functions in response to operational environment. This dissertation proposes a materials-centric framework for robotic embodied intelligence, in which robotic functions arise directly from molecular design. Specifically, embodied energy is realized through materials that store and regulate photochemical energy for controlled propulsion. Embodied control is achieved using shape-morphing materials that encode locomotion and steering via programmable deformation. Furthermore, self-healing is enabled through dynamic covalent polymer networks, restoring functionality and extending lifetime and durability. Together, these studies link molecular design to integrated robotic behavior, advancing multifunctional adaptive materials for intelligent and resilient robotics.

Quantifying the Model Errors Induced by the Traditional Approximation across Weather and Climate Timescales.

Owen Hughes, Climate and Space Sciences and Engineering

Recent work has indicated that a common way that ESMs approximate an important atmospheric force (the ‘Coriolis force’) induces systematic errors in simulations of equatorial atmospheric flow. My thesis removes this so-called Traditional Approximation (TA) from the atmospheric fluid solver in the U.S. Department of Energy’s (DOE) Energy Earth Exascale Earth System Model (E3SM), and ports this model to the National Center for Atmospheric Research’s Community Earth System Model (CESM). The augmented model was implemented and validated to be correct using idealized test cases. Consequently, the model was used to systematically quantify the climatological errors induced by this approximation in operational configurations of CESM and E3SM climate models. Finally, we will use ultra-high-resolution E3SM simulations to quantify the error induced by the TA in simulated extreme weather events, including tropical cyclones and extreme precipitation.

Understanding the Mechanics of Polycrystalline and Granular Materials: Insights from High-Energy Diffraction Microscopy and Graph Theory

Yuefeng Jin, Mechanical Engineering

Graph network theory (GT) provides powerful tools for describing and studying complex particle systems. Polycrystalline and granular materials are two of prime examples of such systems, characterized by intricate microstructures and dynamic local behaviors that give rise to their overall mechanical properties. Despite significant progress, understanding the three-dimensional (3D) mechanics of these materials remains a challenge, particularly in the prediction of microscale mechanics and their relationship to macroscopic properties. High-energy diffraction microscopy (HEDM), an advanced 3D in-situ X-ray technique, enables detailed characterization of materials at the microscale. When combined with GT, HEDM provides new ways of understanding these mechanical behaviors. This dissertation integrates HEDM with GT for two investigations: the deformational behavior of titanium-aluminum alloys, and the system stability of frictional granular materials. These studies focus on understanding localized changes within these systems and their contributions to overall failure mechanisms, bridging the gap between microstructural evolution and macroscopic material response.

Engineering Stability in High-Rate Organic Waste Anaerobic Digestion via Rumen-Inspired Bioreactor Architecture

Renisha Karki (Barbour Scholar), Civil Engineering

A significant portion of organic waste is still landfilled, despite its potential for energy recovery. Anaerobic digestion can convert this waste into bioenergy, but efficient conversion of complex materials typically requires long solids retention times to allow slow-growing microbial communities to fully break down complex compounds and maintain stable performance.

Existing approaches to reduce solids retention times, such as membrane-based systems, can be cost-prohibitive due to fouling and cleaning requirements. My research explores whether principles from the rumen, the cow stomach compartment that enables rapid, stable breakdown of hard-to-digest materials, can offer an alternative. I evaluated a rumen-inspired bioreactor that retains solids while allowing liquids to flow through, enabling high-rate processing by maintaining microbial biomass while increasing throughput. It achieved product yields comparable to high-performing digesters, but at solids retention times up to forty times shorter.

Building on this, ongoing work investigates additional rumen-inspired mechanisms, including controlled micro-aeration, to further enhance stability and sustain high-rate conversion.

Critical Zone Processes within Active Orogens: The Development and Role of the Shallow Subsurface in Landscape Evolution

Sally Keating, Earth and Environmental Sciences

The chemical and mechanical breakdown of bedrock in Earth’s near surface is intrinsically linked to groundwater flow and landscape form, which in turn govern water resources and land surface hazards. The outer veneer of Earth’s crust, or “critical zone,” has thus become the focus of detailed interdisciplinary research, however most works characterize landscapes that are predominantly uniform in time. Steep, rapidly eroding mountains such as the Himalaya differ greatly from these landscapes but remain under-studied. In this dissertation, I develop a novel framework to holistically characterize the chemical and mechanical properties of rock using a suite of field-based methods. I then apply these tools to the central Himalayan Mountains in Nepal to characterize critical zone structure and hydrology across gradients in climate and tectonic history. The novel methodological work and application to catchment scales addresses key, previously unanswered questions about the interplay between hydrology, weathering, and erosion in high mountains.

Decoding Plant–Hydrological Interactions across Climate-responsive Ecosystems: An Integrated Ecohydrological Framework Bridging Field Observations, Physical Processes, and Modeling

Taeho Kim, Environmental Engineering

Water governs ecosystem sustainability by mediating energy and matter transport across the subsurface–plant–atmosphere continuum. Vegetation regulates land–atmosphere water fluxes, yet most ecohydrological models rely on simplified plant parameterizations overlooking heterogeneous plant-level dynamics. This limitation is especially acute in climate-responsive ecosystems, where rapid vegetation change can reshape water and carbon cycling. This dissertation bridges these gaps by integrating long-term field observations, newly identified ecohydrological mechanisms, and advanced modeling frameworks. First, tree sapflow and hydrometeorological measurements classify diurnal sapflow patterns and enable a predictive framework representing tree-level water-use strategies. Second, the study introduces extreme evapotranspiration as an underexplored phenomenon and quantifies its occurrence, drivers, and coupling with ecosystem fluxes. Third, these insights are embedded in an integrated ecohydrological model with generalized particle tracking to quantify water age and subsurface–vegetation connectivity. Finally, AI surrogate models accelerate simulations to robustly evaluate ecosystem resilience under diverse climate scenarios across contrasting ecosystems.

Elucidating Adhesive Interfaces through Sum Frequency Generation Spectroscopy

Gladwin Bryan Labrague, Chemistry

The adhesion strength of a material is dependent on the characteristic of its interface. Many techniques have been used to study interfaces; however, most of these techniques are destructive, not surface/interface sensitive, and/or require highly vacuumed environments. This dissertation utilizes sum frequency generation (SFG) vibrational spectroscopy, which is a non-destructive and sub-monolayer surface/interface sensitive technique that can be operated in ambient conditions. The dissertation research aims to achieve molecular understanding at release coating/substrate, silicone adhesive/substrate, and primer/sealant interfaces to aid in future formulations of adhesive products. The research results fully demonstrated the significance of molecular orientation (standing up or reclined), presence and/or coverage of certain chemical moieties, and molecular interactions (chemical bonding, hydrogen bonding, van der Waals interactions, etc.) at the interface, which substantially impact the performance of an adhesive material, providing important knowledge for future design and development of adhesives with improved properties.

Iron Redox in Silicate Melts at High Pressure: Implications for Deep Earth Redox Evolution

Jiaqi Lu, Earth and Environmental Sciences

Early in Earth’s history, the relatively reduced mantle was extensively molten, forming a magma ocean where redox reactions in iron-bearing silicate melts controlled how oxygen was stored, distributed, and exchanged. Earth’s modern habitability, characterized by an oxidized mantle and atmosphere, is closely related to the redox evolution within the deep Earth. This thesis evaluates a leading hypothesis that iron disproportionation, where ferrous iron (Fe²⁺) separates into metallic iron and ferric iron (Fe³⁺) in magma-ocean melts, contributing to mantle oxidizing as dense metallic iron segregated into Earth’s core. Using Mössbauer spectroscopy, this work establishes a reliable method to determine iron redox state and constrains the high-pressure electronic environment of iron in ultramafic silicate glasses with magma-ocean composition. High-pressure synchrotron melting experiments further define where magma-ocean melts can form. Finally, these constraints were combined to assess whether iron disproportionation in magma-ocean melts can drive large-scale redox evolution, linking deep-Earth processes to the long-term development of surface atmosphere and Earth’s habitability.

Electrochemical Characterization, First Principles & Data-Driven Modeling of Mammalian Cell Transfection via Nanopore Electroporation

Matthew Manion, Chemical Engineering

A central bottleneck to cellular engineering is intracellular delivery of genetic cargos into human immune cells (transfection) with high efficiency, preserved function, and minimal toxicity. My thesis tackles this complex optimization landscape through electrochemical impedance characterization and electrical “feature” extraction to develop a physics-informed, data-driven transfection model for Nanopore electroporation (NanoEP). Therefore, in Aim 1, I characterize how aspects of NanoEP system design (i.e. electrode material, electrochemical reactions, geometry) can affect NanoEP outcomes. Further, I compile data generated from Aim 1 to design a predictive machine learning analysis scheme in Aim 2, approximating the extent of cargo loading and cell health, and extending the predictive ability across cell types. Lastly, Aim 3 utilizes electrochemical and phenotype data to build a more robust NanoEP process prediction system. This work combines computational and experimental methods to elucidate the key system considerations to boost effectiveness and reproducibility of cell transfection for industrial applications.

Multimessenger Constraints on Evolution of Black Hole Scaling Relations

Cayenne Matt, Astronomy and Astrophysics

Numerous pulsar timing arrays have found a gravitational wave background that is too high to be explained by our current models. One possible interpretation of this is that the number of supermassive black holes (SMBHs) in the distant universe is greater than we previously expected, suggesting that the local scaling relations we rely on to predict SMBH mass vary with time. My thesis is determining the details of this evolution. Electromagnetic data alone are insufficient to constrain this evolution, the intricacies of this problem necessitate a multimessenger approach. In my dissertation, I use the semi-analytic binary SMBH population synthesis model, holodeck, to determine the interplay between SMBH galaxy scaling relations, the quasar luminosity function, and gravitational wave background spectrum.

Synthetic Data Generation for Scalable Data-Driven Decision-Making in Smart Manufacturing

Ruo-Syuan Mei, Mechanical Engineering

Data-driven methods, particularly those based on machine learning and artificial intelligence, are transforming decision-making in smart manufacturing. However, the high-quality labeled data they require are often expensive, scarce, and difficult to obtain in real manufacturing settings. My dissertation addresses this challenge by developing synthetic data generation (SDG) approaches that enable more accurate, scalable, and cost-effective manufacturing decision-making. It begins with a systematic literature review and a product-lifecycle framework showing how SDG can support different stages of manufacturing. It also develops a deep-learning-based method for geometric defect detection using synthetic 3D point cloud data. Another part of the dissertation introduces a hybrid SDG pipeline that enables zero-shot vision-based quality inspection on real industrial images, using models trained exclusively on synthetic labeled data. It further extends toward a feedback-driven SDG framework for large-scale production. Overall, this work lowers the data barrier for more reliable, scalable deployment of data-driven solutions in advanced manufacturing.

Spatio-Temporal Dynamics and Causal Drivers of Hydro-Climatic Extremes Under Climate Change

Lisa Nguyen, Applied Physics

Hydro-climatic extremes, including droughts and floods, are intensifying under climate change, posing growing risks to water resources, agriculture, and society. This dissertation investigates spatio-temporal dynamics, causal drivers, and compound impacts of hydro-climatic extremes under a warming climate, with a focus on drought synchronicity, drought dynamics, and extreme precipitation. Integrating complex networks, causal inference, and machine-learning methods, I characterize how droughts synchronize, propagate, and respond to large-scale climate forcing across spatial and temporal scales. Over the U.S., drought synchronicity is strongly modulated by timescale, atmospheric demand, and oceanic variability. Extending globally, I attribute increasing spatial coherence and expanding drought area to anthropogenic warming, amplifying risks to crop production. I will further develop a convolutional neural network framework to track extreme precipitation and drought onset, evolution, and termination, providing a scale-adaptive representation.

Artificial Photosynthesis on III-Nitride Nanostructures

Yuyang Pan, Electrical and Computer Engineering

Gallium nitrides (GaN), the second most produced semiconductors, have shown extraordinary potential for high efficiency artificial photosynthesis, but the carrier dynamics and surface reaction kinetics of nanoscale GaN have remained poorly understood. This thesis investigates the design, fabrication and application of GaN-based nanostructures for high efficiency photo(electro)chemical reactions. Firstly, InGaN/GaN quantum superlattices were developed to exploit quantum-confined Stark effect and the resulting indirect excitonic dynamics, enable one order of magnitude enhancement in the efficiency of solar hydrogen generation under ambient conditions. Furthermore, an integrated heterostructure cocatalyst was rationally designed, constructed and coupled with GaN nanostructures, achieving fast surface reaction kinetics for overall water splitting. Work is currently in progress to apply these newly discovered quantum nanostructures to other challenging chemical reactions, including N2 reduction and oxidation reactions, which have the extraordinary potential to replace the century old Harbor-Bosch and Oswald processes to produce ammonia and nitric acid.

Probing Nanoscale Heat Transport in Non-Reciprocal Systems and Exploring Biometabolism in Small Model Organisms

Kanishka Panda, Mechanical Engineering

Exploring nanoscale heat transport holds the potential to unlock countless possibilities for thermal management and energy conversion while solving outstanding fundamental questions in physical and biological sciences, however, it has remained poorly understood. This thesis aims to unravel two diverse but key avenues of nanoscale thermal science. The primary focus of my thesis is to probe nanoscale heat transport in non-reciprocal systems, exploring novel phenomena such as the existence of heat supercurrents and the photon thermal Hall effect, which remained experimentally inaccessible until now. These discoveries are expected to alter the way we conventionally approach thermal transport and can lead to applications such as dissipation-less nanoscale heat engines and thermal magnetometry. The second part of my thesis involves developing sensitive biocalorimetry to study metabolic heat output in small model organisms and tissues (e.g. Drosophila brains and embryos), aiming to advance the understanding of metabolic disorders and development of therapeutic interventions.

Implantable Biomaterials for Monitoring Breast Cancer Progression and Response to Treatments

Rebecca Pereles, Biomedical Engineering

Triple-negative breast cancer (TNBC) is the most lethal breast cancer subtype due to its ability to spread throughout the body (metastasize) and resist treatments (relapse). To monitor TNBC metastasis and relapse, my lab has developed a biomaterial disk that is implanted subcutaneously, allowing for easy sampling, and mimics sites of metastasis in vital organs, such as the lung. I created a murine model of clinically-relevant TNBC treatments and analyzed cell populations at the disk to generate gene signatures that stratify individuals who would relapse from those who would not. Next, I used the disk in a murine model of obesity to identify hyper-aggressive TNBC. Macrophages and neutrophils were altered within metastatic tissues and may serve as prognostic or therapeutic targets. Finally, I investigated polymeric immune-modifying nanoparticles as a late-stage treatment for TNBC. Nanoparticle administration extended survival compared to traditional treatments by altering the immune cell composition at metastatic sites.

Photochemical, Electrochemical, and Metal-Catalyzed Methods for Aromatic Carbon–Nitrogen Bond Formation and Functionalization

Sabrina Reich, Chemistry

Aromatic carbon–nitrogen (C–N) bonds are ubiquitous in natural products, agrochemicals, pigments, and pharmaceuticals. As such, there is high demand for reactions that form them in a cost-effective and sustainable manner. Traditional methods for C–N bond formation are substrate limited, require forcing conditions (high temperature, long times), and/or employ expensive catalysts. This thesis focuses on overcoming these limitations and thus expanding green methodologies for C–N bond formation. Chapter 1 discusses the limitations of state-of-the-art C–N bond-forming methods. Chapter 2 describes the application of modular in situ photocatalysts for the (CRA)-SNAr pyridination of aryl halides. Chapter 3 details a complementary approach to the same overall transformation that leverages electrochemistry to access electronically diverse N-arylpyridinium products. Finally, Chapter 4 develops transition metal-catalyzed reactions that leverage the pyridinium products from Chapters 2 and 3 as electrophiles in cross-coupling reactions to form C–C, C–O, and C–S bonds.

Dynamical Probes of Cosmology using Galaxy Clusters

Alexander Rodriguez, Astronomy and Astrophysics

Galaxy clusters, the largest gravitationally bound structures in the universe, serve as powerful laboratories for testing cosmological models. This dissertation develops and applies novel dynamical methods to extract cosmological information from galaxy cluster observations. Using the escape velocity of galaxies near cluster boundaries, we construct mass estimates that are independent of assumptions about equilibrium, offering a robust alternative to traditional techniques. We demonstrate that these escape velocity masses agree with independent gravitational lensing measurements. Finally, we extend the framework to constrain cosmological parameters, including the expansion history, directly from cluster data. These results establish galaxy cluster dynamics as a competitive and complementary probe of the universe’s composition and expansion history.

Engineering Dynamic Cell Culture Platforms to Recapitulate Tissue Folding Mechanics

Avinava Roy, Materials Science and Engineering

Biological tissues are inherently dynamic, continuously folding and remodeling, yet in vitro models fail to capture such changes. To bridge this gap, my dissertation focuses on engineering dynamic platforms that mimic tissue shape changes on-demand and capture cell-state changes in real time. Primarily, I have developed two systems: a magnetoactive hydrogel actuated by magnetic fields and a Kirigami-inspired device driven by mechanical actuation. I validated these platforms by modeling airway constriction and arterial tortuosity, showing that dynamic folding is a critical regulator of epithelial barrier function and endothelial health. In my final year, I am utilizing the Kirigami-inspired micromechanical platform to ask a fundamental yet unexplored question: how the extracellular matrix underlying cell monolayers adapts to dynamic curvature. This dissertation provides versatile toolkits for mechanobiology, enabling researchers to move beyond static culture systems and observe cells in dynamic environments.

Atmospheric Controls on Water-Carbon Fluxes and Resilience of Ecosystems

E. Schwartz, Climate and Space Sciences and Engineering

Ecosystems that we rely upon for timber, food, and spiritual connection are jeopardized by a changing climate. Water and carbon fluxes between the atmosphere and land surface are key meteorological processes that impact the climate system. This dissertation investigates the meteorological drivers of surface water and carbon fluxes across seven biomes in North America and the U.S. food system. The first chapter examines the relationship between light availability, moisture, evapotranspiration (ET), and carbon uptake. The second chapter explores the impact of extreme aerosol events on ET and carbon uptake, using the persistent smoke over the upper Midwestern U.S. driven by the 2023 Canadian wildfire season as a case study. The third chapter analyzes how crop yields change in response to fluctuations in light, moisture, and temperature, and suggests opportunities to mitigate crop-specific vulnerabilities through food systems reform.

Flips for Quadrics on Del Pezzo Varieties and Derived Categories for Orthogonal Grassmannian fibrations

Saket Shah, Mathematics

The shapes studied in algebraic geometry, called algebraic varieties, have many interesting invariants and different notions of equivalence. One of these notions of equivalence, called birational equivalence, is closely related to an invariant called the derived category. In this thesis we study how some specific algebraic varieties (called Hilbert schemes of quadrics on del Pezzo varieties) admit special kinds of birational equivalences, called flips; as a consequence we deduce new and interesting decompositions of the derived categories for these shapes. We compare the resulting decompositions to other decompositions, including a new result on decompositions of derived categories for shapes called relative orthogonal Grassmannian.

Reading the Room: How Working-Class Undergraduate Engineering Students Interpret and Navigate the Cultural Expectations of Engineering

Elizabeth Strehl, Engineering Education Research

Title: Reading the Room: How Working-Class Undergraduate Engineering Students Interpret and Navigate the Cultural Expectations of Engineering Abstract: Social class shapes nearly every part of the undergraduate engineering experience, yet it remains one of the least examined identities in engineering education research. Existing work has documented disparities in retention, graduation, and participation using demographic proxies like income and first-generation status, but these measures tell us who is disadvantaged without telling us much about what that disadvantage actually looks and feels like inside an engineering program. This dissertation examines how working-class undergraduate engineering students describe, interpret, and respond to the cultural expectations embedded in engineering education. Engineering carries its own distinct culture, a set of shared norms around academic intensity, professional development, and time commitment, and many of these expectations operate as a hidden curriculum that is not equally accessible to all students. Using a multi-phase, multi-method design at a single research-intensive institution, the study integrates survey data, photovoice-prompted focus groups, and individual interviews to move from mapping the conditions working-class students navigate to examining how they make sense of those conditions over time. The work is guided by the Revised Social Class Worldview Model, which treats social class as dynamic, relational, and interpretive, and centers how students read the economic culture of their institution, understand their own position within it, and experience classism in both structural and interpersonal forms. This project contributes a more detailed understanding of how social class operates within engineering culture and informs institutional efforts to design supports that address the conditions students actually navigate rather than the conditions institutions assume.

Quantifying Current Heterogeneity Among Layered Oxide Battery Cathodes Using Microelectrode Arrays

Wonjoon Suk, Materials Science and Engineering

Correlating the electrochemistry of individual battery particles with global electrode signals provides critical insights into electrode behavior, addressing limitations of operando imaging techniques. Using microelectrode arrays, I designed a platform to quantify current in/out of each particle in an 8-particle electrode, with sub-second and picoampere resolution. I identified abnormal particle-by-particle charging during the first charge, attributed to HF acids in LiPF6 electrolytes that damage particle surfaces. During the initial charge, polycrystalline NMC layered oxide particle cracks due to anisotropic volume change of individual grains; this intergranular cracking exposes undamaged internal surface to enhance particle kinetics. Furthermore, coin cells made with polycrystalline NMC exhibited much faster rate capabilities than single crystal NMC, which do not crack. This highlights that particle cracking can “heal” surface damage caused by HF acids. My findings show how electrolyte chemistry and particle morphology fundamentally impact battery performance, offering new guidance for the design of next-generation batteries.

Removing Emerging Contaminants from Drinking Water

Henry Thurber, Macromolecular Science and Engineering

Emerging contaminants, such as microplastics and per- and polyfluoroalkyl substances (PFAS), are at the forefront of regulation because of their negative impacts to human health. Due to their environmental persistence and not being efficiently removed at water treatment plants, new removal media is needed. First, adhesive-coated substrates were evaluated for removing microplastics. By tailoring the adhesive structure, 99% removal was achieved in 60 minutes. Although the removal was high, the adhesive migrated on the substrate, potentially leaching into water. Utilizing an uncoated stainless-steel mesh in household flow conditions, we achieved 94% microplastic removal in 2.5 minutes and eliminated the adhesive that could contaminate water. Finally, a biobased PFAS adsorbent was developed for drinking water treatment plants. Using functionalized sawdust, PFAS was removed from water at a higher rate and capacity than commercial materials. Overall, we explored new media that was cost effective and could efficiently remove emerging contaminants from water.

Design, Control, and Clinical Validation of Powered, Partial-Assist Ankle Exoskeletons

Katharine Walters, Robotics

Powered ankle exoskeletons have the potential to improve mobility for broad populations. However, current devices are limited by restrictive actuation and unpredictable control. This dissertation addresses these limitations by establishing a comprehensive framework for safe, task-agnostic assistance for both unimpaired individuals and individuals with degenerative joint impairments. A new exoskeleton design leverages quasi-direct drive actuators to overcome traditional trade-offs between torque capacity, compliance, and device mass. A novel trajectory-free, task-agnostic control framework combines ground reaction force amplification with stable, kinematic-based energy-shaping to provide biomimetic assistance, significantly reducing biological ankle torque in unimpaired users across various activities. Extending to individuals with joint impairments, a clinical study with users with chronic ankle osteoarthritis is expected to show pain reductions and improvements in mobility across activities. Further, a pre-clinical validation with users with ankle weakness due to diabetic peripheral neuropathy is expected to show mobility and balance improvements.

L-PBF Additive Manufacturing of Functional Materials to Enhanced Properties

Haozheng Wang, Naval Architecture and Marine Engineering

Additive manufacturing (AM) has gathered the attention since a decade ago, the novel manufacturing technique has revolutionized the complexity of metallic components in designing and fabrication. Components are processed in a layer by layer model throughout to enable the freedom of geometric design. AM also benefits from a reduction to labor costs and enhanced working efficiency with no needs on post processing. AM of structural materials fabrication has reached a high level of maturity owing to its high density and promising mechanical properties, thus building the foundation in the manufacturing community. This novel technique has been applied across multiple engineering fields including automotive, biomedical and aerospace. Since the in application of functional materials such as magnetic materials and thermoelectric materials promote the development of research on novel manufacturing techniques. With the increasing industrial passion on renewable energy applications, for example electric vehicles, the candidate materials of AM have emerged beyond the structural materials to functional materials, particularly magnetic materials and thermoelectric materials where conventional manufacturing techniques have suffered on post processing and manufacturing efficiency. The functional properties of these materials highly replied on the intrinsic correlation of microstructure-processing parameters. Laser powder bed fusion (L-PBF) has been applied on fabricating soft magnetic materials and thermoelectric materials in this prospectus research. The aspect can be split into four sections. 1. Targeting, the material selection can be limited by the nature of L-PBF process, laser focused beam spot carries high energy incidents on the material powder and offers the solid to liquid to solid transition. 2. Optimization, during the extremely complicated melting and solidification process, the resultant microstructure is being affected by multiple factors in various magnitude including fabrication strategies, laser parameters, geometric design and processing environment. 3. Characterization, since microstructure is governing the functional performance, thus it’s crucial to understand the phase formation and solidification mechanism of the targeted materials under certain circumstances and tune the fabrication process to approach the desired functional performance. 4. Application, based on the fundamental research from optimization and characterization, a complex designed components such as EV core can be properly processed with comparable performance.

Calabi-Yau Metrics on Affine Varieties

Ying Wang, Mathematics

We investigate the existence of canonical metrics on affine Calabi–Yau varieties using non-Archimedean techniques. An algebraic variety is Calabi–Yau if it admits a non-vanishing volume form. This feature makes Calabi–Yau at the center of many conjectural theories in mathematical physics, including mirror symmetry, supersymmetry, and string theory. While the existence of canonical metrics on compact Calabi–Yau is well understood, far less is known in the non-compact setting. This dissertation aims to make progress in addressing this gap.

First, we solve a non-Archimedean Monge-Ampère equation on the Berkovich analytification of a complex log Calabi–Yau pair whose dual complex is a standard simplex, answering a question of Collins–Li.

Second, we study the convergence of Archimedean canonical metrics to their non-Archimedean counterparts. This convergence is realized on hybrid spaces, introduced by Boucksom–Jonsson, which encode both Archimedean and non-Archimedean information.

Antigen-Conjugated Scaffolds for Autoimmune Disease Monitoring and Autoreactive T Cell Reprogramming

Sydney Wheeler, Biomedical Engineering

Multiple sclerosis (MS) is a chronic, inflammatory demyelinating condition of the central nervous system (CNS) that is associated with a loss of T cell tolerance. A thorough understanding of T cell dynamics and interactions could improve MS patient care, yet monitoring antigen-specific T cell clones is challenging (<1 in 100,000 in the blood). Moreover, clinical methods to treat MS patients lack antigen specificity and often result in negative side effects and varying efficacy between patients. Our recent work has demonstrated that implantable biomaterial systems can recruit disease-relevant cells in autoimmune conditions, and that if antigens are present, antigen-specific T cells become enriched in these materials. My thesis aims to optimize the fabrication of antigen-conjugated scaffolds and assess the ability of scaffolds to enrich specific clones of T-cells that can be used to study disease development and enable local cell reprogramming as a treatment strategy.

Spatiotemporal Dynamics of Matrix-Mediated Cellular Plasticity in Pulmonary Fibrosis: From Aberrant Angiogenesis to Fibroblast Reversion

Jingyi Xia, Biomedical Engineering

Fibrosis involves profound remodeling of the extracellular matrix driven primarily by activated fibroblasts, but recent studies suggest how other cell types signal to fibroblasts to drive fibrogenesis. This dissertation leverages in vivo and in vitro models to investigate how the two primary cellular constituents of the lung interstitium (ie. fibroblasts and endothelial cells (ECs) communicate following lung injury. I established a multi-scale 3D imaging pipeline to construct a spatiotemporal atlas of murine lung injury and identified an aberrant expansion of non-perfused inflammatory and pro-fibrotic ECs. Next, I developed a novel mouse model that enables in vivo labeling of EC matrix deposition. In parallel, I developed an in vitro model to study how heightened perivascular fiber density destabilizes EC cell-cell adhesions to potentiate TGF-β signaling, drive endothelial-to-mesenchymal transition, and engender a pro-inflammatory phenotype. Collectively, my dissertation describes novel means by which stromal cells regulate the switch between fibrogenesis or resolution.

Embedding Construction Tacit Knowledge into Generative AI for Expert Skill Preservation and Transfer

Gunwoo Yong, Civil Engineering

Construction practice relies heavily on tacit knowledge, often known as know-how, developed through years of hands-on experience. However, such knowledge is difficult to articulate, share, and transfer, limiting its preservation and dissemination. These challenges are further compounded by an aging workforce, increasing the need to support early-career professionals with limited hands-on experience. To address these challenges, my dissertation aims to capture and embed tacit knowledge into generative artificial intelligence (AI), demonstrating its potential to preserve and transfer expert skills by supporting two expert roles. Specifically, one research branch aims to identify ergonomic problems and solutions by guiding generative AI to focus on workers’ poses, reflecting ergonomic experts’ know-how, while the other aims to generate schedules by enabling generative AI to embed schedulers’ experiential knowledge of construction activity sequences. Together, these research branches contribute to reducing reliance on tacit knowledge isolated in individuals, supporting expert skill preservation and transfer in construction.

Social-Ecological Processes Underpinning Human-Wildlife Coexistence

Anna Yue Yu (Barbour Scholar), Environment and Sustainability

Wild animals provide both vital benefits (like seed dispersal) and costs (like crop destruction) to human societies. Governing these tradeoffs is key to sustainable human-wildlife coexistence, a pressing global challenge as people and wildlife share nearly two-thirds of Earth’s land. My dissertation applies an interdisciplinary framework and tests novel hypotheses to address critical gaps in understanding the social-ecological processes facilitating coexistence. At the global scale, I survey conservation professionals worldwide to identify trends, drivers, and monitoring gaps in coexistence across regions and taxa. At the local scale, I draw on a case study from the Tibetan Plateau, a global biodiversity and cultural hotspot, to examine Tibetan communities’ perceptions of risks from an expanding bear population and management preferences under unique social contexts. Integrating wildlife ecology, socio-psychological theory, and community-based approaches, my results generate global patterns and context-specific, actionable insights for managing recovering carnivores and fostering sustainable coexistence in shared landscapes.

Understanding, Assessing, and Measuring Patient Perspectives of Connection in Care

Jennifer Zamudio, Industrial and Operations Engineering

Human Factors/Ergonomics (HF/E) can aid our understanding of human interaction with different elements of the healthcare system, to enhance safety and well-being of both patients and doctors. Research suggests that establishing a meaningful connection during clinical encounters is as important to doctors as it is to patients, enabling better health outcomes and less engagement in risky-behaviors leading to non-adherence. However, the underlying dimensions of this connection remain poorly defined and fragmented. If HF/E engineers are to redesign environments, workflows, and technologies to support meaningful encounters within constraints, we must understand how patients conceptualize connection and develop novel approaches to quantify these interactions. This dissertation uses a mixed-methods approach leveraging qualitative, survey, and computational semantic methodologies to examine prior theoretical dimensions of connection, developing a multi-dimensional scale of connection, and assessing the feasibility of using recurrence analysis to track patterns in engagement in these dimensions during conversation.

Harnessing Low-Dimensionality for Interpretable, Controllable, and Scalable Diffusion Models.

Huijie Zhang, Electrical and Computer Engineering

Diffusion models have sparked a revolution, yet their theoretical foundations remain underdeveloped, creating challenges for generalization, controllability, and efficiency. This dissertation addresses these gaps with a central hypothesis: diffusion models succeed because they uncover and exploit low-dimensional structures within high-dimensional data. Building on this idea, the work develops theoretical insights that drive practical advances across these core challenges. For generalization, it examines how diffusion models trained on finite datasets move beyond memorization to efficiently learn low-dimensional distributions. For controllability, the discovery of a low-rank structure enables the first local editing method and a robust watermarking technique by manipulating within the low-dimensional subspaces. For efficiency, this low-dimensional structure leads to major computational gains, including methods that achieve state-of-the-art image generation in a single step. By addressing these issues, this research advances next-generation generative AI models that require minimal data and computation while producing high-quality, interpretable, and trustworthy outputs.

Revealing Unconventional Charge Density Wave States in Three-Dimensional Systems by Nonlinear Optics

Weizhe Zhang, Physics

Charge density wave (CDW) is a long-studied collective electronic phenomenon characterized by periodic charge modulation that breaks certain symmetries, reconstructs the electronic band structure, and induces distinct transport properties. Recent studies have revealed that novel CDW phases, such as ferroaxial and chiral CDW, further break inversion reflection and rotation operational symmetry and enable unconventional emergent phenomena. Studying the CDW systems from a symmetry-based perspective therefore provides a crucial insight into these nontraditional CDW states and their associated properties. In this thesis, we present our studies of the ferroaxial CDW in rare-earth tritelluride (RTe3) system, chiral CDW structure in EuAl4, and vortex CDW texture in BaTiS3 using temperature-dependent second and third harmonic generation (SHG and THG) techniques. By tracking symmetry changes across phase transitions, we provide an advanced understanding of the evolution of crystal structures in CDW states. Our projects demonstrate the power of nonlinear optical techniques in unveiling hidden symmetry features and highlight their potential in unraveling the complex relationships between CDW and the resulting phenomena.

Towards Safe and Human-Aligned Reinforcement Learning: Theories, Algorithms and Applications

Qining Zhang, Electrical and Computer Engineering

Reinforcement learning (RL) provides a powerful framework for decision-making and underpins recent advances in language models, online platforms, and healthcare support. However, in these applications, the most informative signals ultimately come from humans: through explicit preferences, implicit behavioral outcomes, and safety considerations. These signals are abstract and often expensive to collect, creating fundamental challenges for the design and analysis of RL algorithms intended for high-stakes, real-world deployment.

This dissertation advances RL with human feedback along three directions. First, it develops sample-efficient Q-learning and policy-optimization algorithms that operate directly on explicit human preferences, enabling reliable policy learning under general and even unknown preference models. Second, it studies the fundamental limits of human-centric online experiments and develops RL frameworks that incorporate implicit human behavioral feedback, accounting for constraints, costs, commitments, and multiple performance metrics. Third, it applies these principles to time-critical, safety-sensitive applications, including pandemic interventions and lung cancer screening.