UC Riverside

Open-source projects are widely reused in commercial software, yet its collabora-tive nature exposes it to significant security challenges, particularly N-day vulnerabilities. These vulnerabilities remain exploitable after patches have been released, largely due to delayed patch propagation in decentralized ecosystems. This research addresses the criti- cal issue of prolonged vulnerability exposure by exploring the underlying causes of patch delays and developing automated tools that can help accelerate the patch porting process and reduce the window for attackers. We first present a comprehensive measurement study of the Android kernel patch ecosystem, which systematically analyzes how security patches move from the Linux main- line through various layers of customization by chipset manufacturers and OEM vendors. Our findings indicate that patch delays are a systemic issue, with some patches taking months—or even over a year—to fully reach end-users, which increases the risk of exploita- tion. We analyzed the underlying causes, and one significant reason is that maintainers lack knowledge about which versions are affected by vulnerabilities. In other words, they are unsure when a vulnerability was introduced and which versions are impacted, making it unclear whether the versions they maintain need to be patched. Based on the above observations, we need to speed up the patch porting process to reduce the attack window of N-day vulnerabilities. Identifying the affected versions of these vulnerabilities is crucial for the patch porting process. Therefore, we tackle the challenge of bug bisection—the process of tracing vulnerabilities back to their originating commits. Tra- ditional methods, such as dynamic testing and heuristic-based BIC (bug-inducing-commit, the change that first introduced the vulnerability into the codebase) identification, have shown limitations due to environmental inconsistencies and oversimplified assumptions. To overcome these issues, we introduce a novel approach that uses under-constrained sym- bolic execution to analyze code statically across multiple versions. This method precisely identifies whether the vulnerability logic exists in a given version, thereby isolating the bug-inducing commit. However, the above method still faces several limitations. It requires a proof- of-concept, supports only a narrow range of bug types, and its accuracy is not very high (although it is higher than that of traditional methods). These shortcomings drive us to ex- plore alternative approaches. Finally, we enhance bug bisection by employing large language models (LLMs) that combine code diffs and contextual commit messages. This multi-step filtering approach, which uses both coarse-grained and fine-grained analysis, significantly improves the accuracy of vulnerability detection. Together, these integrated techniques can help accelerate the patching process and reduce the exposure window for N-day vulner- abilities, contributing to a more secure open-source ecosystem. These contributions offer practical solutions for swiftly mitigating vulnerabilities, enhancing open-source security, and ensuring robust resilience in critical software systems.

Augmented reality (AR) continues to evolve with applications across fields such as entertainment, education, and public safety. As we envision the future of a rich AR ecosystem with world-scale AR and collaborative interactions, the demand for responsiveness in user experience becomes more challenging. In this work, I focus on the different aspects of responsiveness in different AR scenarios. In the setting of a single-user, world-scale AR environment, I propose a 3D model retrieval framework that makes intelligent decisions to reduce the communication latency of transferring models from an edge server. When multiple users are present, I introduce an automatic synthesis of a coordination protocol that enables low-latency coordination of virtual objects between users, while respecting real-world spatial constraints.Furthermore, the rapid growth of new 3D content data representations, such as Neural Radiance Fields (NeRF) and 3D Gaussian Splatting (3DGS), has significantly expanded the potential for creating photorealistic scenes within the mixed reality domain. However, the massive data size of 3DGS poses challenges for efficient content delivery. To address this, I propose an optimized framework for scene delivery through customized, layered 3DGS scenes combined with intelligent scheduling algorithms, ensuring efficient and high-fidelity 3D content distributions. This work contributes to advancing AR by addressing latency, synchronization, and content delivery challenges, paving the way for seamless, immersive, and collaborative AR experiences.

Electromigration (EM) has emerged as a critical reliability concern in modernvery-large-scale integration (VLSI) systems, especially as technology scales down to the nanometer regime. The increasing current densities and the miniaturization of interconnect dimensions exacerbate EM-induced aging and failure effects. Additionally, Joule heating, resulting from high current densities, introduces significant temperature gradients along interconnects, leading to thermomigration (TM) effects that can rival or even surpass EM forces. Traditional EM models, such as Black's equation and the Blech limit, often neglect these spatial temperature gradients and are limited to single wire segments, rendering them inadequate for accurate reliability analysis of complex multi-segment interconnect structures in advanced integrated circuits (ICs). This dissertation addresses these challenges by developing advanced EM analysis methodologies that explicitly incorporate the effects of Joule heating-induced temperature gradients in multi-segment interconnects. The key contributions of this work are threefold:

1. Electromigration Immortality Check Considering Joule Heating Effects:We propose a novel voltage-based EM immortality check method for multi-segment interconnects that accounts for spatial temperature gradients due to Joule heating. By deriving an analytical solution for the steady-state EM-TM stress distribution, we extend the traditional voltage-based immortality criteria to include TM effects. This method enables more accurate and less conservative EM sign-off analysis by capturing the interplay between electrical and thermal effects in interconnects. Numerical results on IBM power grids and synthesized power delivery networks demonstrate that the proposed method significantly improves accuracy compared to state-of-the-art EM immortality checks. 2. Fast Electromigration Stress Analysis Considering Spatial Joule Heating Effects: We develop a fast numerical solution for transient EM-induced stress analysis in multi-segment interconnect trees, explicitly considering TM effects. We transform the coupled EM-TM partial differential equations (PDEs) into linear time-invariant ordinary differential equations (ODEs) and apply an extended Krylov subspace-based model order reduction technique. This approach significantly reduces computational complexity, enabling efficient time-domain simulations for both void nucleation and growth phases under time-varying currents and non-uniform temperature distribu- tions. The proposed method achieves up to 28x speedup over existing semi-analytic methods for interconnects with up to 1000 branches, with negligible loss of accuracy. 3. Coupled Electromigration and IR Drop Analysis Considering Spatial Joule Heating Effects: We propose a new TM-aware EM tool, an enhanced simulation tool that couples EM analysis with IR drop calculations while explicitly incorporating TM effects due to Joule heating. Building upon the existing EMSpice framework, the proposed tool integrates TM effects into all key analysis steps, from immortality checks to hydrostatic stress calculations. The tool employs model order reduction techniques to accelerate simulations, achieving an average of 22% reduction in computational time for power grids up to 256 x 256 nodes. Notably, we observe that Joule heating can lead to void nucleation along interconnect branches, not just at cathode nodes, and can reduce maximum stress levels and void sizes. These findings indicate that TM effects can mitigate EM-induced failures in certain scenarios. Validation against finite element method simulations confirms the accuracy of the proposed tool, highlighting its effectiveness for practical reliability and sign-off analysis of on-chip power grids. The methodologies developed in this dissertation provide a comprehensive frame- work for EM analysis that accounts for the combined effects of electromigration and ther- momigration in multi-segment interconnects. By incorporating spatial temperature gradi- ents due to Joule heating, the proposed approaches offer enhanced accuracy and efficiency for reliability assessment in advanced VLSI systems. These contributions have significant implications for the design and validation of reliable integrated circuits, enabling better utilization of interconnect reliability limits and facilitating the development of more robust electronic systems.

As modern computing systems evolve, the emergence of diverse memory technologies and hardware accelerators presents significant challenges and opportunities. With the DDR standard facing density challenges and the emergence of non-volatile memory technologies such as Cross-Point, phase-change memory, and fast FLASH media, compute and memory vendors are contending with a paradigm shift in the datacenter space. The decades-long status quo of designing servers with DRAM technology as an exclusive memory solution is likely coming to an end. Future systems will increasingly employ tiered memory architectures (TMAs), in which multiple memory technologies work together to satisfy applications’ ever-growing demands for more memory, lower latency, and greater bandwidth. Concurrently, hardware accelerators have computing capabilities for applications in a broad spectrum of domains. However, from the operating system’s (OS) perspective, the entrenched execution models for each accelerator differentiate them as second-class citizens from the host and segregate per-device multitasking schemes locally without global control.This thesis explores two fundamental aspects of heterogeneous computing: (1) comprehensive methods for memory-access collection in TMAs and (2) the promotion of accelerators to first-class citizens in a heterogeneous computing system by treating them equally as host-level devices in a virtually unified preemptive multitasking environment. To address TMA challenges, this thesis evaluates various memory-access collection methods and proposes a hybrid tiered-memory management strategy that enhances system-wide visibility and performance. For accelerator elevation, this thesis presents Acceleratory Rights (AccelRights). AccelRights uses intermediate representation (IR) as an abstraction representing interchangeable scheduling entities across all devices, regardless of their diverse instruction set architectures (ISAs), and proposes hardware-independent context switching and task migration mechanisms. The proposed methodologies demonstrate significant improvements in memory efficiency, system-wide computational performance, energy consumption, and carbon footprint. By bridging memory hierarchy optimization with efficient accelerator management, this research contributes to the development of next-generation heterogeneous computing systems.

In many different biological systems, the migration and distribution of specific cells, such as neural stem cells within the brain, has a large influence on the dynamics of the whole system. Furthermore, biological experiments are often costly and time consuming to perform and are unable to test certain hypotheses. For this reason, mathematical modeling is an invaluable tool which is able to make predictions as well as isolate specific variables in order to test and confirm a given biological hypothesis. In this thesis, we examine cell migration in three different biological systems. In each system, we develop a novel model based on underlying mechanistic assumptions and then use this model to test and investigate previous biological hypotheses. The modeling approaches in this thesis range from micro to macroscale models.

In the first section, a computational model of Pseudomonas aeruginosa migration is introduced and used to explore the impacts of bacterial reversals and wrap mode on efficiency of motion both with and without chemotaxis in different environments. Model simulations demonstrate that wrap mode can increase diffusion of P. Aeruginosa in a non-chemotactic environment. Wrap mode is shown to not only increase the chemotactic efficiency but also to reduce the metabolic expenditure of the bacteria that is required for migration. The experimentally observed frequency of wrap mode was used to demonstrate the existence of an optimal trade off between chemotactic efficiency and metabolic cost. It was also shown that the wrap mode can allow the bacteria to move along the Fungi and increases exploration of multimodal chemotactic profiles. Namely, respondents of P. aeruginosa to the chemotactic signals produced by a hypothetical fungal network were studied in a liquid environment. For cells undergoing a run-reverse pattern, it is shown that the bacteria are likely to remain in the first local chemoattractant maximal production site on a fungi that they find. With wrap mode, however, the bacteria are more easily able to escape these local maximums to further explore their neighboring environment along the hyphae. Finally, we find that the dispersal of bacteria moving along a Fungi on agar is decreased when the network density of the fungi increases.

In the second section, we build an agent-based model of the migration of therapeutic neural stem cells in rodent brain. In the naïve mouse brain, the developed model based on generalized q-sampling imaging is shown to better account for the tissue microstructure over models based on diffusion tensor imaging. In calibrating to recent experimental data, we find that the model is able to fit the quantitative statistics of in-vivo experiments, however struggles to match the levels of distribution to the olfactory bulb seen in the experimental data. In addition, we show that both the quantitative and qualitative statistics are sensitive to the injection location. This highlights the need for additional experimental data. If distribution to the olfactory bulb is a persistent phenomenon, it suggests that future models of therapeutic neural stem cells in the naïve brain may need to include other forces such as chemotaxis or blood flow. If it is not a persistent phenomenon, it suggests that the experimental results may be a product of the injection location.

In the final section, we build a multiphase model of blood clot deformation under fluid flow. We explain some of the recent experimental literature that demonstrates the importance of red blood cells on the clot stability and then create a multiphase model which includes red blood cells, fibrin network, less and highly activated platelets, and platelet-rich plasma. The model couples together both macroscopic and microscopic forces and accounts for pairwise surface tensions between all phases.

Over the last two decades, the universe of insect taste has significantly expanded, from the initial identification of receptors expressed on peripheral neurons to the elucidation of complete neuronal circuits governing memory, locomotor output, homeostasis, and numerous behaviors associated with feeding. In parallel, neurobiologists have leveraged the genetic workhorse Drosophila melanogaster in both its larval and adult form to pursue these directions, with each developmental stage affording unique advantages as a model system for dissecting taste. Our work includes efforts to optimize the process of behavioral data analysis in adults, where we develop an adaptable pipeline for high-resolution analyses of multiple features associated with feeding on liquid food sources. Using this approach and established choice behavior assays, we identify the regulation of appetitive tastant feeding via pharyngeal gustatory receptor neuron (GRN) populations, and specifically a subset of pharyngeal GRNs that express sugar receptor Gr43a. However, how taste sensing and feeding behavior is shaped across metamorphosis is less understood. To better understand the taste system’s influence on behavior across development, we first developed a model for larval tastant exposure that permits us to assay adult behavior using non-toxic amounts of bitter tastants. We identified that exposure to certain tastants as larvae imbued attenuated avoidance to innately aversive tastants as adults across behavior assays such as food choice and proboscis extension responses. This shift in behavior was specifically linked to the identity of the tastant encountered during larval development. Additionally, we observed that behavioral modification required both functional bitter taste and intact mushroom body and dopaminergic neuron activity, where gustatory memory is formed and stored. Our results suggest that attenuation of avoidance to innately bitter compounds may require multiple levels of putative taste circuits, from the periphery to central processing components. Interestingly, silencing of dopaminergic neurons implicated in learned avoidance seemed to potentiate avoidance behavior, revealing the required regulation of learned aversion pathways in habituating avoidance. Overall, this work represents the first genetic and circuit-wide dissection of how a tastant response may be modulated across development following exposure during early life.

The two major citrus postharvest decays green mold and sour rot of citrus are caused by Penicillium digitatum and Geotrichum citri-aurantii, respectively, and result in significant crop losses in California citrus if not properly managed. Several postharvest fungicides are registered to manage green mold, including imazalil, thiabendazole, azoxystrobin, fludioxonil, pyrimethanil, propiconazole, and natamycin, but only the latter two are highly effective against sour rot. Cyproconazole is a demethylation inhibitor (DMI) triazole fungicide currently undergoing registration for citrus postharvest use. My results on vitro toxicity and efficacy for management of these decays provide support for its registration, particularly in the presence of DMI resistance in sub-populations of the causal pathogen with EC50¬ values > 0.5 µg/ml. Cyproconazole was effective in reducing mycelial growth of P. digitatum isolates sensitive or resistant to imazalil, and EC50¬ values ranged between 0.05 and 0.90 µg/ml. For G. citri-aurantii isolates sensitive or moderately-resistant to propiconazole, EC50 values ranged between 0.10 and 0.83 µg/ml. Inoculated citrus fruits treated in laboratory or experimental packingline studies with cyproconazole alone or in mixtures with other registered fungicides reduced decay by sensitive or resistant isolates of both pathogens by 78 to 99% compared to untreated controls. Cyproconazole was compatible with storage and packing fruit coatings when applied as mixtures using a low-volume spray system. Cyproconazole was compatible with inorganic salts like sodium carbonate and sodium bicarbonate in high-volume applications, and with sanitizers like sodium hypochlorite and peroxyacetic acid without reducing efficacy. The combined in vitro sensitivity and efficacy studies demonstrate incomplete cross resistance between cyproconazole and the registered DMIs imazalil and propiconazole that can improve efficacy in integrated pest management programs in the presence of resistant pathogens. In G. citri-aurantii, point mutations in the DMI target sites of CYP51A result in moderate or high resistance to propiconazole without conferring fitness penalty costs, whereas mutations in CYP51B do not have an effect. The genes encoding the two target molecules were identified and characterized using whole genome sequencing and classical molecular cloning techniques, and they can be distinguished using allele-specific diagnostic PCR primers. Lastly, G. candidum was identified as being a weak pathogen of citrus, causing sour rot on senescent fruit with compromised host defenses.

Urban water demand is surging due to increased population and climate change effects on freshwater reserves. In arid places, the water demand surpasses the supply, putting cities under water-stress conditions. On the other hand, there is an increase in demand for green cities that incorporate significant green spaces for ecosystem benefits such as recreation, cooling, and carbon sequestration. However, sustaining the vegetation’s health requires irrigation, costing a substantial amount of water. Therefore, this dissertation aimed to address the knowledge gaps on the trade-offs between water conservation and vegetation ecosystem services, develop predictive models to enhance irrigation decisions, and assess vegetation responses to irrigation water conservation policies at a regional scale using a combination of field measurements and secondary data. Field experiments of two warm-season turfgrass species, Buffalograss, and St. Augustinegrass, were established, and six irrigation rates and two irrigation frequency treatments were applied for two to three years (2021-2023). Plant visual quality and temperature were monitored using handheld sensors and an Unmanned Air Vehicle (UAV). Then, CO2 efflux, soil moisture, temperature, carbon isotope discrimination (Δ), and carbon content in soil and plant biomass were measured. Lastly, satellite data on vegetation, precipitation and air temperature, household water use, and income data for Southern California during the 2013-2017 drought were integrated and analyzed. The results showed that reducing irrigation diminished turfgrass visual quality, cooling potential, and CO2 efflux of both turfgrass species but at different rates among species due to their physiological differences. Reducing irrigation also negatively correlated with the Δ in Buffalograss (r=-0.45) and St. Augustinegrass (r=-0.11) biomass, indicating plant water stress. However, a positive (r=0.4 at 10 cm depth) correlation was found between irrigation rate and Soil Organic Carbon (SOC) in St. Augustinegrass, showing increased carbon input. Furthermore, models based on UAV data showed potential for predicting soil moisture and CO2 efflux. Lastly, results showed that mandatory water conservation measures in southern California led to a 26% reduction in water use and a 5% Fraction Vegetation Cover (FVC), where Low-income groups had a 9% lower baseline FVC than High-income groups but a higher reduction and less rebound in FVC.

This dissertation examines the construction of trans identity within and beyond systemic structures that both validate and constrain legibility. Through close readings of transmasculine memoirs and interwoven personal narrative of my own, this project interrogates the ways trans authors navigate the paradox of seeking recognition while resisting the frameworks that render them pathologized subjects. While trans narratives offer a means of self-articulation and cultural visibility, they are also shaped by institutions that define trans identity through medicalization, correction, and coherence. By tracing the genealogy of transmasculine memoirs from the 1990s to the present, this project highlights the ways these texts reinforce and resist dominant narratives of transition, dysphoria, and institutional validation.Drawing from José Muñoz’s disidentification, Leslie Feinberg’s freedom of gender expression, Sianne Ngai’s ugly feelings, Cameron Awkward-Rich’s trans maladjustment, Hilde Lindemann’s holding and letting go, and Joseph Slaughter’s bildungsroman as a technology of personhood, this work challenges the expectation that trans narratives must resolve in legibility and wholeness. Instead, it advocates for reading trans stories through their narrative ruptures, contradictions, and refusals to conform to coherent identity scripts. The first chapters situate trans memoirs within a literary and historical context, demonstrating how their structure creates space for trans visibility while also reinforcing and/or navigating the exclusions of medical and legal frameworks. Later chapters analyze contemporary trans narratives that disrupt the bildungsroman model through aesthetic nervousness, genre subversion, and spiritual reclamation. At its core, this project critiques how trans people are made to author themselves into coherence for systemic validation. It argues that liberation lies not in perfecting legibility but in resisting the compulsion to be rendered fully knowable by institutions that legislate existence. Reading across trans narratives—both those that uphold and those that fracture traditional identity scripts—this dissertation calls for a trans storytelling/reading praxis that resists closure, embraces healing, and makes space for identities of becoming.

This dissertation presents a suite of novel methodologies designed to advance multi-user augmented reality (AR) systems by addressing challenges in localization, mapping, and real-time collaboration. Key contributions focus on enhancing visual-inertial odometry (VIO) and introducing infrastructure-less cooperative SLAM techniques.

Firstly, a Point-Line Cooperative Visual-Inertial Odometry (PL-CVIO) framework is proposed to improve localization accuracy, particularly in low-feature environments. By integrating point and line features and enabling feature sharing between neighboring robots, PL-CVIO leverages geometric constraints to achieve robust, cooperative localization. The framework employs covariance intersection (CI) to ensure consistent state estimation across multiple agents.

Secondly, a novel map-assisted VIO system is introduced by leveraging Neural Radiance Fields (NeRF) to encode compact and photorealistic 3D maps. These maps provide robust geometric constraints for localization, addressing key challenges such as pose initialization, drift correction, and environmental adaptability. A pose initialization model is proposed by using geodesic errors. Besides, an online VIO algorithm is developed, which leverages both real-world and NeRF-rendered images to update the state, demonstrating significant improvements in accuracy and robustness.

Thirdly, we propose CooperSLAM, a lightweight, infrastructure-free cooperative SLAM algorithm designed for multi-user AR in dynamic and resource-limited environments. CooperSLAM enables efficient peer-to-peer communication and sparse map feature sharing, enhancing scalability while reducing bandwidth requirements. By decoupling map points and key frames and introducing opportunistic relocalization strategies, CooperSLAM facilitates effective collaboration without reliance on centralized infrastructure.

Extensive simulations and real-world experiments validate the performance of the proposed methods. Results demonstrate substantial improvements in localization accuracy, robustness, and scalability compared to existing methods. This work contributes to the development of intelligent, collaborative AR systems designed to function effectively in dynamic and infrastructure-less environments, offering potential applications in immersive technologies, robotics, and related fields.