UC Riverside

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 mechanisms of magnesium oxide nanoparticles (nMgO) and magnesium hydroxide nanoparticles (nMg(OH)2) were investigated within the Huinan Hannah Liu Research Group at the University of California, Riverside. Bioresorbable nMgO has been investigated for antimicrobial effects, but the antimicrobial effects of nMg(OH)2 have been largely ignored. Here, we identified the antibacterial effects of nMg(OH)2 using previously reported methodology in Escherichia coli, Pseudomonas aeruginosa, Staphylococcus epidermidis, Staphylococcus aureus, and Methicillin-resistant Staphylococcus aureus to produce comparable results across studies. In addition, the effects of nMgO and nMg(OH)2 were established in logarithmic phase bacteria to determine bacteriostatic or bactericidal reductions in cell viability, if adaptation to repeated nanoparticle exposure occurs, and putatively determined the mechanisms of nMgO and nMg(OH)2 antibacterial activity. Importantly, we identified a reduction of mature bacterial biofilms and a reduction of cell viability in stationary-phase planktonic bacteria that was not equivalent to the minimum inhibitory or minimum bactericidal concentrations identified in Lag phase bacteria. The outcome of this study provides scientific knowledge previously unknown to the scientific community and may be applied to downstream materials and bioengineering research that may be applied in clinical settings for 1) their potential synergistic effects with antibiotics, and 2) reduce the development of antibiotic-resistant bacterial populations with a reduction in the overall use of antibiotics.

We present Adaptive Skip, an efficient inference-time acceleration technique for large language models that dynamically skips LLM layers using reinforcement learning. Our method formulates layer skipping as a decision-making process where a lightweight policy network determines which transformer layers to compute or skip based on input complexity. Unlike previous approaches, Adaptive Skip maintains full compatibility with KV caching by using consistent skip patterns across tokens and also does not require any retraining of the base LLM. We train the policy using Group Relative Policy Optimization with a multiplicative reward function that jointly optimizes output quality and computational efficiency. Experiments on summarization tasks demonstrate significant speedups (19.4% on CNN/DailyMail and 76.9% on Reddit TLDR) while preserving output quality within 2% of the full model. Adaptive Skip outperforms existing layer skipping methods by learning more effective layer selection patterns adaptive to the input without modifying the underlying language model. Our approach provides a foundation for generalizing to diverse computation modes beyond binary layer skipping.

School-based problem-solving teams (SB PSTs) are widely used in K-12 public education to address student academic, behavioral, or social-emotional concerns through evidence-based problem-solving procedures and interventions. While research suggests SB PSTs can be effective, little is known about how they function in practice without researcher oversight or what factors contribute to better team or student outcomes. Drawing from team science, scholars have called for a more standardized framework to evaluate the SB PST processes, and Rosenfield et al. (2018) proposed using the Input-Mediator-Output-Input (IMOI) framework for this purpose. This study will survey SB PST members to examine relationships between team inputs (team ability/skill and administrator support), mediators (affective states, cognitive states, and behavioral processes), and outputs (team satisfaction, student outcomes, and systemic outcomes) using structural equation modeling. This study will also examine if there are statistically significant differences between reported team outcomes and student or systemic outcomes. Findings will contribute to a more systematic understanding of SB PST functioning and effectiveness.

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.

Plant microbiomes have been shown to facilitate a wealth of benefits to their plant hosts. Many of these benefits are mediated by the secondary metabolites these microorganisms produce. Certain microbial secondary metabolites have been shown to directly inhibit the growth of pathogenic microorganisms that infect and damage crop plants. Currently, there is a desperate need for inhibitory compounds to combat the devasting disease of Citrus plants, Huanglongbing. This bacterial phytopathogen has put the domestic citrus industry in peril. Current short-term solutions are unsustainable, and long-term solutions remain many years from commercial availability. In this dissertation, we utilize a high throughput culture and screening platform to probe the microbiome of field grown Citrus and search for native bacteria able to inhibit the growth of the associated causal agent of Huanglongbing. Utilizing an array based microbial cultivation system, the Prospector®, we created an isolate collection of over 1,800 isolates recovered from above ground and below ground Citrus plant tissue. This isolate collection was screened against the closest culturable relative to the causal agent of Huanglongbing, Liberibacter crescens, to identify potential sources of inhibition. This screen highlighted 35 isolate morphotypes that showed consistent inhibition of L. crescens in a dual microbial assay. These isolates were further characterized through genome sequencing. Several isolates are previously undescribed, having no species level matches to the Genome Taxonomy Database. We screened metabolome extracts of the isolate collection against L. crescens. Several extracts produced highly inhibitory extracts while others, though inhibitory in a dual microbial assay, did not. This was attributed to secondary metabolite regulation and was further indicated by evidence of siderophore production in most isolates, which is tightly regulated. We show that a siderophore predicted to be produced in the citrus microbiome displayed growth inhibition of L. crescens. In an isolate producing an inhibitory metabolite extract, we identify the antibiotic compound phenazine-1-carboxamide present in an inhibitory metabolome fraction. Utilizing an ex vivo hairy root assay, we report promising preliminary results suggesting this compound may be active against a pathogenic Liberibacter species. This work demonstrates the citrus microbiome as a rich source of Liberibacter antagonism.

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.

Inhibitory control (IC) is the capacity to interrupt an action in order to reach a specific goal. Impulsivity is the tendency to act rashly despite potentially negative consequences. Conceptually, they imply an inverse relationship, but this has not been consistently found in previous research. IC is measured using performance-based tasks, while impulsivity is generally measured using self-report questionnaires, and this format difference has led to issues in previous studies when comparing directly. In chapter one, this problem is addressed by conducting an Exploratory Factor Analysis (EFA) to identify how performance-based measures of IC and questionnaire scores of impulsivity correlate and group together. We identified four factors across 19 total measures of IC and impulsivity. Three factors consisted of measures with significant loadings from impulsivity assessments, while the fourth factor was showing significant loadings from IC tasks, suggesting IC and impulsivity may be separate constructs driven by separate underlying processes. In chapter two, to explore the relationship structure of IC and impulsivity assessments in a novel way, three network analyses were conducted, using 1) IC measures, 2) impulsivity measures, and 3) both IC and impulsivity measures with data from a healthy adult sample. These analyses revealed sub-networks, or “communities,” that were also largely dominated by assessment type, though some overlap across IC and impulsivity was observed in the full model. Chapter three compares a novel, gamified cognitive task based on the traditional cancellation task, with other traditional IC tasks. We found that UCancellation RT-based metrics significantly predicted TOVA RT variability, suggesting its possible utility as a more appealing alternative to the TOVA in certain cases. Ultimately, the results from this dissertation could help inspire future researchers to remove the redundancy of assessments used to measure IC and impulsivity in both research and clinical settings, while also introducing a novel, gamified measure of IC that may serve as a useful alternative to less-engaging traditional cognitive tasks.

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.

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.