UC Irvine

Modeling and simulation is an important part of the process of designing embedded systems. It can be used to identify the effectiveness of new hardware architectures, and analyze performance trade offs. In this dissertation, we model a new CPU architecture called the Grid of Processing Cells (GPC). The GPC architecture is cache-less and is based on a grid/arraycomputing platform, with a primary focus on reducing main memory contention. Among the many constraints and limitations of today’s computing devices, one of the biggest open problems is the memory bottleneck. Observed already in the first single-processor architectures, the single lane through which data and instruction streams are transferred to and from the main memory impedes the traffic flow and often results in heavy congestion, limiting the processing speed severely. This problem is only multiplied in today’s shared-memory multi- and many-core architectures and, despite sophisticated multi-level cache hierarchies, remains as a grand challenge that stands in the way of efficient parallel processing. While Moore’s law is coming to an end due to the physical limits on further increasing clock frequencies, the demand for parallel computing will only increase with the advance of big data and deep learning applications. Realizing the required computing performance with efficient processor architectures requires novel parallel platforms that do not suffer from a central memory bottleneck (and ideally do not need any costly cache hierarchies).

The consequences of climate change present significant challenges revolving around plant resilience, food systems, and training the next generation of problem solvers. My dissertation approaches these challenges with a lens of plant health, human health, and the educational training to help create a more sustainable future. In Chapter 1 of my dissertation, I examine how the avocado (Persea americana), a subtropical fruit with increasing production demands, uses water and responds to climate variability. I examined various plant functional traits of five commercially relevant avocado cultivars, spanning from least intentionally bred to most intentionally bred: Lyon, Hass, Fuerte, GEM, and BL516. I found that cultivars differ in their leaf- and stem- water management strategies and climate responses, with insight that newer cultivars may be better suited to a rapidly changing climate, particularly for trees grown in rainfed conditions. Building on these findings, in Chapter 2, I explored the socio-ecological impacts of avocado production in Michoacán, Mexico, the largest avocado-producing region in the world. Specifically, I investigated how avocado expansion affects ecosystems and the occupational health of production workers. From this analysis, I developed a potential knowledge-action model aimed at promoting long-term health of humans and ecosystems in the face of continued agricultural demand. Beyond the agricultural system, in Chapter 3, I investigated how a local invasive grass, Bromus diandrus, interacts with microbial communities, focusing on growth and early ontogeny. We established a soil microbial inoculum gradient, and found that plant response to higher density microbial conditions influenced growth, but was constrained by an allometric relationship between roots and shoots; lower density inoculum resulted in plants with greater shoot-than-root growth, restricting total biomass production through seedling establishment. Finally, in Chapter 4, I worked alongside Ridge 2 Reef colleagues to develop strategies for improving training and modernizing STEM graduate education programs to better equip the workforce for addressing grand challenges, including those related to climate change and the future of agricultural systems. We developed five design principles to best prepare future problem solvers to tackle complex, interdisciplinary challenges, collaboratively. Together, the projects of my dissertation integrate knowledge from multiple disciplines to address ongoing and emerging challenges of sustainability of food systems in the face of global change.

Nanoscale magnetic tunnel junctions (MTJs) have become transformative components in advanced spintronic technologies, offering unprecedented capabilities due to their large magnetoresistance and efficient spin-transfer torque. MTJs are the key building blocks of nonvolatile magnetic memory, tunable nanoscale microwave sources, high-sensitivity microwave detectors, true random number generators, and spintronic neurons. Beyond their practical uses, MTJs provide a unique platform for exploring nonlinear magnetic dynamics in confined geometries, driving forward fundamental discoveries in the field.

This dissertation investigates the utilization of MTJs to study novel regimes of magnon tunneling and to develop an ultrafast spintronic microwave frequency tracker. Magnetic anisotropy in the MTJs nanomagnets is tuned by varying the nanomagnet thickness, enabling precise tailoring of the quantized magnon spectrum. This approach optimizes resonant microwave rectification efficiency, facilitating the design and realization of ultrasensitive spintronic microwave detectors. Moreover, pairs of MTJs tuned to distinct resonance frequencies are employed to create ultrafast microwave frequency trackers operating within the UHF band (1–3 GHz) with sub-microsecond response times. A compact, standalone X-band frequency tracker integrating MTJ technology with digital electronics is also developed. These MTJ-based detectors and trackers are radiation-hard and exhibit significantly higher sensitivity than industry-standard semiconductor counterparts, such as Schottky diode-based devices.

The study of spin-transfer torque-driven magnetization dynamics in these MTJs has led to the discovery of a novel magneto-dynamic phenomenon: nonlinear magnon tunneling. In this process, a magnon in one nanomagnet of the MTJ tunnels across the nonmagnetic barrier and splits into a pair of magnons in the other nanomagnet. The nonlinear hybridization arising from this coupling is examined as a function of the external magnetic field, with strong resonant coupling observed at discrete field values.

This work advances the understanding of nonlinear dynamics in magnetic nanostructures and demonstrates the potential of MTJ-based technology in the realization of novel, ultrafast spintronic microwave devices.

Language model agents are tackling challenging tasks from embodied planning to web navigation to programming. These models are a powerful artifact of natural language processing research that are being applied to interactive environments traditionally reserved for reinforcement learning. However, many environments are not natively expressed in language, resulting in poor alignment between language representations and true states and actions. Additionally, while language models are generally capable, their biases from pretraining can be unaligned with specific environment dynamics. In this dissertation, I cover our research into rectifying these issues through methods such as: (1) mapping high-level language model plans to low-level actions, (2) optimizing language model agent inputs using reinforcement learning, and (3) in-context policy improvement for continual task adaptation.

A comprehensive, data-driven methodology for analyzing aircraft performance factors frompublicly available operational radar and weather data is presented. This approach offers a versatile framework for assessing issues such as fuel burn, correlating operations with noise events, and gaining insights into airspace dynamics to support the integration of future air- craft concepts. One key application of this methodology is the evaluation of operational and environmental factors, along with their interactions, that contribute to noise variations across different aircraft types—a significant and growing challenge for airports. Aviation-induced community noise remains a persistent issue, and this methodology is driven by the need for operational adjustments to mitigate its effects. By identifying the sources and factors influencing noise propagation, informed modifications to flight procedures can be imple- mented. The methodology uses publicly available, ADS-B historical surveillance data from the Opensky Network and classifies departures from arrivals. It integrates noise monitoring recordings from the airport ground noise monitoring network with flight trajectories and collects weather data at the airport and noise monitor positions during specific timestamps using the NOAA Rapid Refresh (NOAA RAP) model. The methodology was applied to operational flight data from Seattle-Tacoma and Boston-Logan airports, leveraging several years of ADS-B data and noise monitor recordings for Airbus A319, A320, A321, Boeing 737-700, -800, -900, and Boeing 777-200LR/ER aircraft. Noise variations were analyzed as a function of observed parameters—including aircraft type, flight trajectory, airline, wind, temperature, pressure, and relative humidity—and inferred variables, such as aircraft config- uration, thrust, and weight. Spearman correlation matrices and gradient-boosting decision tree models were employed to evaluate the impact of these factors on noise variation. Other applications implementing this methodology include estimating fuel consumption by tracking weather along flight paths, assessing the feasibility of Advanced Air Mobility (AAM) oper- ations through visualizing current airspace organization, and developing speed and altitude profiles for various aircraft types.

Channel Charting (CC) is an emerging technique in wireless communications that aims to map spatial relationships between user equipment (UEs) using Channel State Information (CSI). Unlike conventional localization techniques, CC operates in an unsupervised manner, relying solely on CSI without external positioning data. This thesis explores various algorithms such as MUSIC, Bartlett, MVDR, Minimum Norm, Linear Regression, and ISQ, for estimating Angle of Arrival (AoA) θ and the distance d, forming the foundation for CC. Through extensive simulations, we evaluate the accuracy, computational efficiency, and robustness of these algorithms under different wireless channel conditions, including Line-of-Sight (LOS), Quasi Line-of-Sight(QLOS) and Quasi Non-Line-of-Sight (QNLOS). Our results indicate that high-resolution spectral estimation techniques like MUSIC provide superior accuracy but at a higher computational cost. In contrast, simpler approaches such as Linear Regression and ISQ offer computational efficiency but with reduced precision in complex wireless environments.Additionally, we examine the impact of varying antenna configurations and UE densities, highlighting the tradeoffs between performance and resource constraints. The insights gained from this study contribute to the optimization of CC techniques, paving the way for more efficient and scalable wireless localization systems. Future research may further enhance these methods by integrating machine learning or hybrid models to balance computational efficiency and accuracy.

This study investigates the use of conventional angle of arrival (AoA) estimation algorithms, including Bartlett, Minimum Variance Distortion Response (MVDR or Capon), FFT, and Minimum Norm, for estimating θ (AoA). These methods are evaluated alongside a previous approach, the inverse of the root sum squares of channel coefficients (ISQ), linear regression (LR), and a novel application of the MUSIC algorithm for estimating the distance from the base station ρ in the context of channel charting. The evaluation focuses on the visual quality of channel charts, dimensionality reduction performance metrics such as trustworthiness (TW) and connectivity (CT), as well as the execution time of these algorithms. The results show that while Bartlett, MVDR, FFT, and Minimum Norm demonstrate comparable performance to previously studied techniques, the Minimum Norm algorithm exhibits significantly higher computational complexity. Furthermore, it was observed that the MUSIC algorithm achieves very high performance when estimating both θ and ρ. This study further quantifies the performance of the Bartlett algorithm for estimating both θ and ρ, providing insights into its efficiency and effectiveness compared to the MUSIC-based approach.

Huntington’s disease (HD) is a devastating autosomal dominant neurodegenerative disease that is characterized clinically by psychiatric disturbances, cognitive dysfunction, and loss of motor control. The disease is caused by an expanded CAG repeat in exon 1 of the HTT gene that results in an expanded glutamine tract in the HTT protein. In addition to severe striatal degeneration, cortical atrophy is a key pathology and reduced cortico-striatal functional connectivity has been defined in HD patient’s years before clinical onset(Unschuld et al., 2012). Stem cell derived neurons provide insights into early dysfunction that may inform optimal therapeutic intervention. Several groups, including our own, have shown using human pluripotent stem cells (hPSC’s) differentiated towards striatal and cortical neuronal lineages, that HD lines display phenotypic and molecular alterations at multiple stages of differentiation. While these studies provide valuable insight into potential disease mechanisms, they primarily involve a snapshot from a single assay in multiple patient lines and variability between cell lines and differentiation methods can make the data difficult to interpret and reproduce. In Chapter 1 we used an embryonic stem cell isogenic series(Ruzo et al., 2018), engineered to express a range of CAG repeat lengths or HTT knockout (KO), to generate integrated cell signatures for pluripotent stem cells and differentiated cortical neurons. By integrating multiple omics datasets, e.g. transcriptomics, epigenomics, and proteomics, into coherent biological pathways to generate hypotheses about network level dysfunction we provide a baseline for further validation and perturbation. These networks associated with CAG expansion were then compared to those generated for total HTT loss to understand the contribution of HTT loss of function compared to gain of function given that the HD mutation results in both loss of normal HTT function as well as gain of aberrant toxic functions. We used this comparative analysis to gain insight into the extent to which each of these contribute to disease progression.

Neuroinflammation and activated microglia are also implicated in disease progression. Microglial activation correlates with severity of striatal neuron loss in post-mortem HD brains, and PET imaging also indicates an elevation of inflammatory markers in pre-clinical patient brains. Expression of mutant HTT (mHTT) in mouse microglia results in a cell-autonomous upregulation of pro-inflammatory gene expression. In addition to this innate inflammatory phenotype, functional dysregulation in HD microglia may have deleterious consequences in the HD brain. HD patient blood monocytes are hyper-reactive and display impairments in phagocytosis and chemotaxis. Multiple studies have reported that, similar to patient monocytes, human HD iPSC derived microglia are hyper-reactive to LPS stimulation and have elevated toxic reactive oxygen species release; however, functional deficits have not consistently been observed. There remains a major gap in defining molecular and functional responses in iPSC-derived microglia like cells (iMG). In Chapter 2 we use patient-derived as well as CRISPR modified isogenic iPS cell lines differentiated into microglia to determine how mHTT expression affects homeostatic microglial functions including phagocytosis and chemotaxis. Similar to previously reported studies, we were unable to reproducibly detect impairments in homeostatic microglial function. We also examined the effects mHTT expressing microglia on iPSC derived MSNs in co-culture. We investigated the effect on inhibitory synapse number in the MSNs by immunocytochemistry and electrical activity using the Axion microelectrode array. We did not observe any changes in synapse number or firing rate for MSNs as a result of co-culture with mHTT expressing microglia.

The effects of mHTT expression on early disease progression may be more subtle than what can be detected using the functional assays of Chapter 2. In Chapter 3 we focused first on determining if there is an altered transcriptome in human iPSC derived mHTT expressing microglia (iMGs). While significant cell line variation was observed, upregulation of classical immune response and autoimmune associates’ pathways were observed for each HD patient line. We examined whether this cell autonomous inflammatory transcriptome exerts an effect upon an unaffected mouse brain using single nuclei sequencing of MITRGxFIRE mouse brains transplanted with expanded CAG and control iMGs. Variation was observed based on the cell line transplanted and the sex of the animal that likely reflects differences observed in HD mouse models and patients. Despite these differences, our data indicates that mHTT expressing iMGs affect transcriptional differences in pathways associated with synaptic vesicle release and with apoptosis in MSNs. In turn, synaptic plasticity may be negatively affected in cortical neurons. Additionally, we observe widespread downregulation of genes associated with primary cilia, organelles that are important for relaying environmental signals to the cell.

Biomaterials continue to be at the forefront of medical advancements, offering a broad range of functionalities which can be tailored to a variety of needs. The efficacy of an implanted biomaterial is dependent on its ability to mitigate the foreign body response (FBR). The FBR is characterized by a cascade of cellular recruitment and signaling, leading to implant encapsulation and persistent inflammation. Consequently, advancements in biomaterials design often focus on developing their immunomodulatory capability. This capacity to direct local immune cell behavior can arise from the structural design of the material, exploiting known relationships between induced cell characteristics and downstream expression. Porous structures are of particular interest in these applications due to their superior capacity for cell infiltration and biomimicry of native tissues. Specifically, previous work from this group demonstrated that a polymeric biomaterial known as a bicontinuous interfacially jammed emulsion gel (bijel)-templated material (BTM) caused a significant reduction in fibrosis and increase in vascularization in an initial in vivo investigation. Those immune benefits were hypothesized to be caused by the uniform pore structure and negative Gaussian curvature characteristic of the BTM. This dissertation presents a body of work which describes direct relationships between BTM morphology and immune cell behavior via both computational and in vitro avenues. Computational modeling of generic actin-ratcheting cells interacting with simulated BTMs and particle-templated materials (PTMs) revealed significant differences in cell shape and migration behavior between the substrates. Deeper analysis suggested that the locally high curvature in regions of the PTM significantly deterred migration, in contrast to the uniform curvature of the BTM which did not similarly impede motility. In vitro examination of macrophages interacting with these substrates revealed expected differences in cell shape, correlated to lower inflammatory expression in BTMs. Fibroblasts seeded into PTMs and BTMs in vitro had, in addition to corresponding shape differences and phenotype benefits, markedly different migration behaviors, in line with the modeling predictions. These findings contribute key insights to the fundamental understanding of cell behaviors in porous substrates and inform future development of novel immunomodulatory biomaterials.

Intelligent Transportation Systems (ITS) increasingly depend on sophisticated sensor fusion techniques to accurately track vehicles and optimize traffic flow. However, these systems are vulnerable to malicious attacks targeting physical sensors, particularly inductive loop detectors (ILDs), commonly used for vehicle detection at intersections. This thesis addresses the critical problem of understanding how sensor spoofing attacks, specifically magnetic loop spoofing on ILDs, impact the reliability and effectiveness of ITS. To analyze these vulnerabilities, this study employed a comprehensive simulation framework replicating a realistic multi-sensor intersection control environment. Data was gathered through simulated scenarios involving a combination of ILD and camera-based detection under varying weather conditions—including clear, heavy rain, and heavy fog—to mimic diverse operational circumstances. An ILD spoofing attack was modeled, injecting false vehicle detections and suppressing real detections to evaluate the system's resilience. The analysis leveraged standard multi-object tracking metrics such as Multiple Object Tracking Accuracy (MOTA), ID Precision, ID Recall, and ID switches, alongside overall intersection throughput, to quantify the attack's impact. The findings revealed significant performance degradation under attack conditions. Specifically, metrics such as MOTA dropped by approximately 25% and IDF1 decreased by around 20%, indicating substantial performance degradation. Intersection throughput also suffered notably, decreasing by up to 30% under high-intensity attack scenarios, further illustrating the tangible impact on traffic efficiency. Additionally, false positives increased by nearly 35%, and ID switches rose significantly, further emphasizing the attack’s disruptive capability. Furthermore, this research proposes avenues for future work, including extending analyses to other sensor modalities like cameras and radar, and developing multi-layered defense mechanisms to bolster resilience against coordinated sensor attacks. Ultimately, this thesis contributes to the intersection of cybersecurity and transportation engineering, guiding the development of secure, reliable, and resilient ITS infrastructure.