UC San Diego

The plight of students experiencing homelessness in the United States is reflected in bothstudent outcome data and in those personal narratives shared of their experiences. Those student experiences and the reasons for homelessness are often quite complex and layered. Data reveal students experiencing homelessness consistently underperform their housed peers in many academic indicators and are twice as likely to experience a mental health challenge. Further, inequitable school outcomes result in a greater chance of experiencing attendance and behavioral concerns. These outcomes also result in a lower high school graduation rate at 67.8% which impacts the ability to access a college experience. This study explores the impact of school programs and services on the success of high school students experiencing homelessness in public education. Review of literature conveys the impact of federal policy in removing barriers for students experiencing homelessness and the correlation between a positive school climate and the academic achievement of this student group. Moreover, the literature provides how homelessness impacts socio-emotional development and how schools can be a supportive factor in strengthening this development. Finally, the literature suggests the mitigating role school- based programs and school staff can have on the educational experience of students experiencing homelessness. Keywords: cultural proficiency, education equity, education leadership, homeless students, McKinney-Vento, poverty

Wearable electronics have revolutionized human-machine interfaces and healthcare monitoring by enabling non-invasive, continuous, and real-time assessment of motion and physiological signals. This dissertation presents advancements in these fields through the development of novel stretchable devices and the integration of machine-learning techniques in four chapters. The first chapter introduces a wearable human-machine interface that integrates an inertial measurement unit (IMU) and electromyography (EMG) sensors. Leveraging a deep-learning neural network trained on a composite dataset, the system effectively mitigates diverse motion artifacts, enabling continuous and precise robotic control in highly dynamic environments. The second chapter presents a fully integrated, single-transducer echomyography (EcMG) system for long-term, wireless muscle signal monitoring. Designed to be worn on the forearm, the system employs a deep-learning model to correlate single-transducer signals from forearm muscles with hand gestures, achieving continuous and precise tracking of 13 hand joints with a mean error of 7.9°. Furthermore, the biomedical application of the EcMG system for real-time respiratory monitoring will be explored, demonstrating its capability to track diaphragm movement in both patients and healthy subjects. The third chapter discusses a photoacoustic wearable patch that integrates an array of ultrasonic transducers and vertical-cavity surface-emitting laser (VCSEL) diodes on a common soft substrate. The high-power VCSEL diodes generate laser pulses capable of penetrating more than 2 cm into biological tissues, activating hemoglobin molecules to produce acoustic waves. These waves are subsequently detected by the transducers, enabling high-resolution three-dimensional temperature imaging of hemoglobin. Finally, the fourth chapter introduces a stretchable capacitive sensing device with integrated electrodes and control electronics, offering enhanced signal quality. By utilizing a dielectric calcium copper titanate oxide (CCTO) as the adhesive layer, the device achieves increased electrode capacitance, resulting in an improved signal-to-noise ratio in the acquired biopotential signal.

Turbulent mixing in the ocean surface boundary layer (OSBL) mediates the transfer of energy, momentum, and gases between the atmosphere and the ocean. While the dynamics that drive these exchanges happen on small lateral and vertical scales relative to the size of the ocean, they are crucial in setting the behavior of Earth’s climate on global scales. Historically, small-scale processes near the ocean’s surface have been difficult to observe. Using novel measurements of the OSBL enabled by advances in instrumentation and multi-platform observational techniques, this thesis investigates specific OSBL turbulent dynamics. The onset and growth of Langmuir circulations (LCs) is observed from simultaneous airborne infrared imagery of sea surface temperature and subsurface in-situ measurements. For weak, fetch-limited wind wave forcing with stabilizing buoyancy forcing, LCs appear non-uniformly in space. During a period of LC growth and diurnal warm layer (DWL) deepening, subsurface temperature structures show temperature intrusions into the base of the DWL of the same scale as bubble entrainment depth during the deepening period. A large-eddy simulation run with observed initial conditions and forcing reproduces the onset and rate of DWL deepening, but exhibits coherent temperature structures with a larger aspect ratio than in observations, with large sensitivity to the numerical representation of surface radiative heating. At the base of the mixed layer, a drifting thermistor chain is used to observe temperature fluctuations consistent with turbulence in a stratified shear layer resulting from a mixture of Kelvin-Helmholtz and Holmboe instabilities. The size and frequency of these structures depends on the surface forcing regime, defined by the balance of wind-shear, wave-shear, and convective turbulent kinetic energy production. Thorpe scale estimates of dissipation and entrainment are consistent with the observed rate of mixed layer deepening, while the outer vertical scale of the turbulent region is correlated with the wind forcing magnitude. Below the mixed layer, observations of velocity from an array of drifting profiling instruments are used to relate spatial gradients in near-inertial wave energy flux to array-scale lateral vorticity gradients. These unique observational studies can aid in improving future numerical simulations and parametrizations used in global climate simulations.

Industrial development increasingly pollutes agricultural soil and groundwater with heavy metals such as arsenic (As) and cadmium (Cd), which contribute to accumulating heavy metals in staple crops such as rice. Prior studies have investigated heavy metal uptake in plants but have not addressed how to decrease heavy metal accumulation and toxicity in Oryza sativa (rice). Herein, we attempt to prevent heavy metal toxification in edible and aerial tissues of rice by overexpressing root-targeted promoter wheat phytochelatin synthase 1 (TaPCS1) and heavy metal associated3 (AtHMA3). Wild-type and transgenic rice was grown in a hydroponic culture with elevated concentrations of As and Cd. ICP-MS analysis observed an average increase of 68.8-fold, 0.6-fold, 66-fold, and 2.8-fold in TaPCS1 As roots, As shoots, Cd roots, and Cd shoots, respectively, compared to the wild-type. Moreover, we observed a 13.7-fold concentration increase of Cd in AtHMA3 root tissue and a 31.5-fold increase of Cd in the edible shoot tissue compared to the wild type. We conclude that TaPCS1 and AtHMA3 might not play a role in the production of phytochelatin that sequesters As and Cd into the roots of the plants, reducing toxicity in the shoots. Furthermore, future analysis of different overexpressing root-targeted promoters may yield a potential solution to lowering known carcinogens of As and Cd accumulation in rice.

Mind the Gap is an 84-minute, movie-length collection of songs that explores the relationship between truth and truthiness, and the ability of music to reclaim lived experience and trauma through fictionalized, narrative sound. This dissertation explores the process and content of this work.

The growing prevalence of data analysis methods, including modern machine learning, has increased interest in applying these techniques to sensitive personal data. However, data analysis on sensitive data poses different privacy risks depending on the type of data and the way it is collected and analyzed. This necessitates privacy-preserving data analysis to be tailored to each specific context in order to protect individual privacy properly while providing insightful knowledge. To address this challenge, this dissertation focuses on three interrelated aspects of privacy-preserving data analysis: (i) identifying the privacy model, (ii) crafting the formal definition of privacy, and (iii) designing the mechanisms satisfying the privacy definition. By examining these aspects, the dissertation demonstrates how to design practical privacy-preserving data analysis across various applications, ranging from traditional linear queries to large language models.

This dissertation proposes a behavioral coaching framework for adults with autism, focusing on head orientation behavior in triadic (three-way) conversations. Triadic interactions, common in both social and professional settings, can be particularly challenging to navigate as they require balanced attention distribution between conversational parties for effective communication. Research indicates that individuals with autism are more likely to exhibit non-normative orientation patterns in multi-party interactions. To analyze these patterns, we developed novel body and head orientation estimation models that process low-resolution 3D point cloud data from LiDAR sensors. The body orientation model uses ellipse fitting, while the head orientation model employs an ensemble of neural network regressors trained on hand-crafted geometric features. Our models achieve mean absolute estimation errors of 5.2 degrees for body orientation and 13.7 degrees for head orientation. Using our orientation estimation system, we designed a triadic conversation setup where adults with autism engaged with two interviewers in a casual conversation. We demonstrated statistically significant differences in attention distribution behavior between autistic and non-autistic individuals in these triadic interactions. Building on these findings, we designed a behavioral intervention program to provide personalized feedback on head orientation behavior to individuals with autism. The intervention consisted of data analysis from an initial conversation session, followed by video modeling and constructive discussions about specific head orientation patterns. Among the 11 participants who received the intervention, 10 demonstrated statistically significant improvements in a post-intervention session. Finally, we enhanced the human-guided intervention by employing artificial intelligence (AI) techniques to imitate the decisions of human coaches. We developed a multivariate time-series classifier to detect social exclusion behavior using sequences of estimated head orientations. Our Time-Series Transformer model detects instances of social exclusion behavior with a precision of 85.7%. This dissertation outlines a road-map for designing an AI-based, end-to-end behavioral coaching system. Chapter 1 introduces a novel orientation estimation system which lays the foundation for analyzing orientation behavior. Chapter 2 applies this system within a human-guided intervention program. Finally, Chapter 3 integrates AI-driven decision-making, representing the initial steps towards an automated behavioral coaching framework.

“Who you are is how you lead.” — Dare to Lead, Brené BrownMy stage management journey began at Takarazua Stage Company in Japan. Takarazua employs a strict, hierarchical system where responsibilities are clearly defined, and production needs take precedence over everything and everyone else. In that environment, I learned to value directness and honesty, as feedback was given bluntly, focusing solely on achieving the best results as efficiently as possible. This approach became an intrinsic part of my identity. When I started my graduate studies at UC San Diego, I encountered a leadership model centered on empathy and individuality, where company members had diverse working styles that felt unfamiliar to me. I feared my previous style would be too harsh for some people and ineffective outside its original cultural context. At the same time, abandoning the structural working method entirely to adopt a new approach felt like erasing a part of myself. During The Rogues’ Trial, a production with thirty cast members, I faced numerous challenges, including scheduling conflicts and the sudden illness of a lead actor, requiring three understudies to step in at the last minute. Feeling overwhelmed and unsure how to proceed, I turned to my core values of directness and honesty. I openly shared my struggles with the company and asked for their support. The response was remarkable. The company members rallied behind the process, offering me their full support. Even when I had to use a firm voice under pressure, they understood my intent rather than perceiving me as too harsh. Setting clear expectations with honesty, vulnerability, and transparency became the bridge to trust. By fostering trust through open communication and feedback, I discovered a way to serve both the process and the people. Through my journey, I have learned that leadership is not about choosing one approach over the other. It is about intentionally creating my leadership style. The discipline and efficiency I developed in the hierarchical system remain valuable, but those qualities are most effective when paired with trust, openness, and adaptability. I have shaped a leadership style that honors my past while allowing me to continue evolving with new experiences. Trust became the bridge connecting these two worlds, allowing me to lead with clarity and compassion.

Phosphorus is a vital nutrient required for the functioning of living organisms. In marine environments, phosphate is considered the most bioavailable form of phosphorus for microbes. However, phosphate can be scarce, potentially limiting microbial metabolism, energy supply, and ecosystem functions. To overcome phosphate scarcity, microbes produce alkaline phosphatase, a hydrolytic enzyme that enables utilization of dissolved organic phosphorus (DOP) as an alternative phosphorus source. Here, a year-long time series was conducted to analyze alkaline phosphatase activity (APA) at the Ellen Browning Scripps Pier, a high-nutrient coastal environment, to understand the range of factors influencing APA in this ecosystem. APA was observed throughout the majority of the year despite phosphate-replete conditions, a phenomenon referred to as the APA paradox. The APA paradox challenges the traditional role that alkaline phosphatases are thought to play in marine systems. A possible explanation for this finding is that microbes might be using organic carbon liberated from DOP hydrolysis rather than the phosphate. To test this hypothesis, the heterotrophic bacterium Ruegeria pomeroyi was grown on three different DOP compounds as sole carbon sources and APA was measured over 13 days. R. pomeroyi grew on all DOP as carbon sources, but to a lower degree than on glucose. Among DOP sources, maximum growth was observed on glucose-6-phosphate (G6P), followed by adenosine monophosphate (AMP), and adenosine triphosphate (ATP). The highest APA was recorded in the cultures lacking any carbon amendment (-C), with activity levels an order of magnitude higher than in other groups, followed by cultures grown on G6P. These findings suggest that alkaline phosphatases likely play a role in the acquisition of some carbon forms by marine microbes. Overall, this study helps advance the current state of knowledge regarding phosphorus cycling, the APA paradox, and carbon utilization in marine environments.

Artificial intelligence (AI) is primed for revolutionizing cancer treatment. Predictive oncology leverages AI to forecast cancer treatment outcomes for each patient using all available information: genetic, molecular, cellular, and clinical. These methods have evolved into powerful tools for understanding the effects of genetic alterations on treatments. However, the integration of predictive oncology models into routine clinical practice faces significant challenges. This dissertation addresses these challenges.Chapter 1 proposes seven hallmarks for the development, evaluation, and clinical adoption of predictive oncology models. Considerations for each hallmark are discussed, along with an example model scorecard. Subsequent chapters detail several methods that directly address these hallmarks to advance predictive oncology. Chapter 2 introduces a framework to organize protein-protein interaction networks into hierarchical knowledge maps. These maps are highly effective in gaining insights into biological mechanisms, such as treatment response. Chapters 3 and 4 present interpretable or “visible” neural networks (VNNs) to predict response to cancer therapies. To elucidate the underlying mechanisms, VNNs incorporate a hierarchical knowledge map of multi-protein assemblies in cancer, which integrates rare and common alterations across hundreds of clinical panel genes. VNNs for palbociclib and cisplatin identify multiple assemblies that accurately stratify patients as responders or non-responders. Finally, chapter 5 conducts a retrospective evaluation of the palbociclib VNN model to identify patients with durable palbociclib response. Patients with predicted sensitivity to palbociclib had significantly better progression-free survival and overall survival than patients with predicted resistance. Model prediction and biomarker interpretation were integrated to create a personalized report for each patient.