Grants

NINDS - National Institute of Neurological Disorders and Stroke

Project Summary/Abstract The ability to flexibly learn from experience and recall past events depends on the brain’s capacity to adjust synaptic strength in response to activity. This form of synaptic plasticity is not uniformly expressed across a neuron’s dendritic arbor, but instead follows highly compartmentalized rules that enable distinct input streams to be integrated with specificity. Yet, how such compartment-specific plasticity mechanisms are coordinated within and across neurons – and how they translate into meaningful circuit-level output during learning – remains poorly understood. The hippocampal CA3 network, a central hub for memory encoding and retrieval, contains two morphologically and functionally distinct pyramidal neuron subtypes. TE⁺ cells, with their thorny excrescences (TEs), receive dense mossy fiber input from dentate gyrus (DG), whereas TE⁻ cells lack these structures and receive little or no DG drive. How such input- and subtype-specific circuitry implements memory remains unknown. This work integrates multi-scale in vivo imaging, genetic perturbations, transcriptomics, and computational modeling to link synaptic architecture and gene expression to memory computation in heterogeneous CA3 circuits. In the K99 phase, I will examine the synaptic mechanisms of input-specific plasticity using in vivo two-photon glutamate and calcium imaging to monitor compartment-specific dynamics at individual spines during learning (Aim 1.1). Direct in vivo measurement of synaptic weights will guide the development of recurrent neural network models incorporating experimentally derived integration rules (Aim 1.2). Building on these synaptic principles, I will next examine how input-specific mechanisms scale to the cellular and population levels by assessing how DG input drives subtype-specific coding in CA3 using in vivo volumetric calcium imaging (Aim 2.1) and causal genetic manipulation (Aim 2.2). In the R00 phase, I will develop a novel “morpho-tagging” approach to isolate TE⁺ and TE⁻ neurons for transcriptomic profiling (Aim 3.1), followed by in vivo gene perturbation to identify molecular mechanisms that support subtype-specific synaptic plasticity (Aim 3.2). The findings will deepen our understanding of the synaptic, cellular, and molecular mechanisms underlying learning and memory, and may inform strategies to address cognitive dysfunction. I will receive extensive training in experimental and computational methods, particularly in advanced imaging, transcriptomics, and modeling, along with structured career development in scientific writing, leadership, and project management from my mentors. I have also assembled a team of expert collaborators who will provide specialized guidance across all aims. My integrated mentorship team and rich institutional environment will ensure my successful transition to independence and the launch of a multidisciplinary neuroscience program.

NIDDK - National Institute of Diabetes and Digestive and Kidney Diseases

Nocturnin (NOCT) is a NADP(H) phosphatase with highly rhythmic expression peaking in the early dark phase (ZT12). Notably, Noct-/- (Noct-KO) mice are resistant to obesity and hepatic steatosis on a high-fat diet. Given the widespread role of NADP(H) as an essential cofactor in numerous metabolic reactions, it is likely that NOCT has distinct tissue-specific effects, making it difficult to understand NOCT’s role in metabolism using a global KO model. In fact, previous studies have observed contradictory changes in lipid and lipoprotein metabolism across different tissues in Noct-KO mice. Specifically, while previous studies have found that Noct- KOs have delayed chylomicron transit into circulation following olive oil gavage, other studies have found significantly higher circulating VLDL, triglycerides, and cholesterol at certain times of day in the plasma of chow- fed Noct-KO mice as compared to controls. In further contradiction to the lean phenotype seen under high-fat diets, chow-fed Noct-KOs have significantly higher expression of hepatic enzymes involved in fatty acid synthesis during their active phase. Given these seemingly paradoxical phenotypes, this proposal will use novel adipose- , liver-, muscle-, and intestine-specific conditional Noct-KO (cKO) mice, generated via the Cre-loxP system, to better understand NOCT’s role in metabolism. Both male and female cKO and control mice will be subject to a high-fat diet for twenty weeks, with body weight and food intake recorded on a weekly basis, to determine if loss of NOCT in one tissue is sufficient to confer resistance to diet-induced obesity. A separate cohort of cKO mice will be placed in metabolic cages for one week to discern any differences in energy expenditure. At the end of the high-fat diet challenge, body composition measurements will be taken to determine differences in fat and lean mass. Plasma, adipose, liver, muscle, and intestinal tissues will be collected from each cKO line and their respective controls at two circadian phases for further analyses. Plasma lipoprotein classes will be fractionated and characterized to determine how tissue-specific loss of NOCT impacts lipid mobilization and trafficking. Plasma adipokine levels will also be measured to identify any differences in adipocyte metabolism and health. To confirm the mechanism by which any observed phenotype originates, NAD(H) and NADP(H) levels will be measured in each cKO tissue. Further, RNA-seq will be performed on all collected tissues to investigate how tissue-specific loss of NOCT impacts transcriptional networks. Tissues will also be stained with hematoxylin and eosin to define any gross morphological changes. Lastly, lipogenic flux will be measured in both liver-cKO and global KO mice to determine the direct impact on de novo lipogenesis. The experiments outlined in this proposal will define how tissue-specific loss of NOCT, and thus how tissue-specific increases in NADP(H) levels, impact metabolism. More specifically, this proposal will aid in understanding how loss of NOCT protects against diet- induced obesity and hepatic steatosis, which, in turn, can aid in the development of treatments for obesity and obesity-related diseases.

NIAID - National Institute of Allergy and Infectious Diseases

Project Summary This project aims to uncover the transcriptional and metabolic mechanisms that govern the identity, function, and therapeutic potential of mucosal-associated invariant T (MAIT) cells, a non-MHC-restricted, innate-like T cell subset with growing relevance in cancer and infectious disease immunotherapy. By leveraging two clinically relevant molecular glues—CC-647 and DKY-709—that selectively degrade the transcription factors PLZF and Helios, we will precisely manipulate the transcriptional programs of MAIT cells and related innate-like lymphocytes ( T and NK cells) in primary human samples. Using integrated transcriptomic, epigenomic, metabolic, and functional profiling, this study will define how PLZF and Helios sustain MAIT cell effector poise, tissue-homing capacity, and metabolic restraint. The outcomes will enable the rational design of engineered MAIT cells with optimized persistence and functionality for off-the-shelf immunotherapies.

NIAID - National Institute of Allergy and Infectious Diseases

Project Summary/Abstract Shigella species are the cause of bacillary dysentery, a severe diarrheal disease. Shigella are highly infectious, and cause an estimated ~200 million cases and ~200,000 deaths annually, many of which are young children. Because Shigella is a human-specific pathogen, we have—until recently—lacked a genetically tractable and physiologically relevant animal model of shigellosis. As a result, our understanding of shigellosis is based primarily on in vitro studies or non-physiological or non-genetically tractable animal models. Many generally accepted beliefs about Shigella pathogenesis have never been rigorously tested in vivo. In what we consider a major advance for the field, we have developed the first physiological oral infection mouse model for Shigella. Our new model is based on our discovery that a cytosolic innate immune sensor, called the NAIP–NLRC4 inflammasome, is essential for resistance of mice to Shigella. Shigella-infected Nlrc4–/– mice exhibit all the hallmarks of human shigellosis, including weight loss, diarrhea that can be bloody, bacterial replication specifically within intestinal epithelial cells, and a robust neutrophilic inflammatory response. As in humans, shigellosis in Nlrc4–/– mice is self-limiting and resolves spontaneously within a week. Importantly, bacterial virulence determinants known to be essential for virulence in humans are also essential for virulence in our new model. In this proposal, we plan to take advantage of our new Nlrc4–/– mouse model of shigellosis to identify key host and bacterial factors that control pathogenesis. Specifically, our overall hypothesis is that bystander (uninfected) macrophages play a previously unappreciated role in orchestrating immunity to Shigella, and that they do so via two key cytokine axes: a TNF axis that promotes death and expulsion specifically of infected epithelial cells, and an IL-12 axis that promotes IFN gamma production. To test our overall hypothesis, we propose three Specific Aims: (1) Test the hypothesis that macrophages are essential to resist Shigella; (2) Test the hypothesis that TNF protects against Shigella by inducing the selective apoptosis and expulsion of infected intestinal epithelial cells; (3) Test the hypothesis that a macrophage-IL12-lymphocyte- IFN gamma circuit acts to close the epithelial cell niche to Shigella. Our new model of Shigella pathogenesis involves many players—e.g., macrophages, TLRs, IL-12, IFN gamma, TNF—that are not currently believed to play major roles in resistance to Shigella. Thus, we expect that results with our rigorous in vivo experimental model will significantly revise our understanding of this important infectious disease.

National Institutes of Health

Dissemination and Implementation Research in Health (R01 Clinical Trial Optional)

National Institutes of Health

Dissemination and Implementation Research in Health (R03 Clinical Trial Not Allowed)

National Institutes of Health

Dissemination and Implementation Research in Health (R21 Clinical Trial Optional)

NIDCD - National Institute on Deafness and Other Communication Disorders

The “Dissemination and Implementation Science Travel Award and Conference Event” (DISTAnCE) is designed to build scientific capacity to accelerate the uptake of Dissemination and Implementation (D&I) research across NIDCD’s mission areas (i.e., hearing, balance, taste, smell, voice, speech, and language). D&I science aims to narrow the gap between scientific discovery and its application in health care and educational settings to promote the adoption and integration of evidence-based practices, interventions, and policies. The research-to-practice gap has been durable and presents a significant challenge to furthering evidence-based practices for many disciplines. For example, despite the considerable evidence base in communication disorders (ASHA n.d.), its clinical uptake has been disappointing (Douglas et al., 2022b). This gap negatively impacts the health and education of millions of individuals with communication disorders (NIDCD n.d.). Several factors have been identified that impede scientific progress in D&I (NIDCD, 2023), but primarily, there is insufficient scientific capacity in D&I to support noticeable growth. The need for mentoring support and cultivation of the D&I research community have been recognized as being critical to advancing D&I research in NIDCD’s mission areas (Douglas et al., 2022a, NIDCD, 2023). The DISTAnCE program aims to increase scientific capacity and address unmet needs in D&I across NIDCD’s mission areas by supporting 12 awardees and 6 mentors annually to attend (1) the “Conference on the Science of Dissemination and Implementation in Health” (CSDIH) and (2) the new DISTAnCE Workshop, which will immediately follow the CSDIH. The DISTAnCE Workshop will focus on topics highly relevant to NIDCD’s mission areas such as identifying and engaging stakeholders and choosing appropriate D&I frameworks, strategies, outcomes, and funding mechanisms (e.g., Douglas et al., 2022a, NIDCD, 2023). Critically, half of the Workshop time will be spent in mentoring periods wherein awardees work with their mentors to refine their draft grant proposals that incorporate D&I aims and methods. Through in-person discussions, group functions, post-workshop virtual meetings, and a dedicated online community, DISTAnCE will strengthen the developing D&I research community. Within the year following the DISTAnCE Workshop, awardees will broadly disseminate information learned at CSDIH and DISTAnCE through presentations, seminars, articles, and other broad dissemination tactics. Outcomes data regarding awardees’ success in securing funding for grant proposals that includes D&I aims and methods will be collected at 3-years and 6-years following their participation. Over the 5-year funding cycle, DISTAnCE aims to support the funding success of 60 promising scientists focused on D&I research within NIDCD’s mission areas and to promote broad dissemination about the value of D&I to nurture the pipeline of future D&I researchers and clinical implementation partners.

OD - NIH Office of the Director

PROJECT SUMMARY/ABSTRACT Advances in computer vision-based biomechanical analysis and human pose estimation have helped expand access to quantitative methods for movement analysis. These advancements have highlighted the utility of markerless motion capture (MMC) for overcoming the logistical challenges presented by marker-based systems. For example, MMC eliminates the cumbersome and time-consuming process of marker placement, and MMC systems can capture details that cannot be obtained by marker-based systems. However, commercially available MMC systems often fail to meet clinicians’ and researchers’ needs, and the proprietary nature of these systems prevents researchers from personalizing the systems and addressing the issues specific to their environment. Our team has developed MMC acquisition software designed to integrate into clinical workflows, as well as cutting-edge algorithms for biomechanical analysis, which are supporting several NIH-funded projects. Our collaborators using our software already report its many strengths, which speaks to the gaps we are filling, including whole-body tracking of arms and hands. While we have successfully deployed this software in multiple NIH-funded research labs, there are still barriers to making the system broadly accessible. For example, its implementation requires on-site engineering support. Additionally, output datasets and metadata are not readily shareable due to the fact that they contain identifiable information from subjects, such as raw videos, so additional processing steps are required for de-identification of data. The overall objective of this Research Software Engineer R50 proposal is to apply best software practices to this code and transform our current video acquisition and biomechanical analysis software, which works well for our research projects with experienced engineering support, into a high-quality open-source product that is easy to deploy in new labs and supports FAIR data sharing and reproducible re-analysis of exported datasets. Specifically, we will ensure our software is truly accessible to a broad research and clinical audience by developing a robust, user-friendly, easy-to-deploy, open-source, and FAIR4RS markerless motion capture acquisition system and biomechanical analysis pipeline (Aim 1). Further, we will facilitate researchers’ ability to share and access movement data by implementing a data layout that produces de-identified, FAIR- compliant datasets and metadata that can be easily imported and exported (Aim 2). We anticipate that these improvements will democratize access to MMC by providing researchers and clinicians a robust and reliable software resource that can be easily implemented across a diverse range of settings, and this software will improve data sharing by producing datasets that can easily be imported to and exported from repositories for broad dissemination.

NIMH - National Institute of Mental Health

7. Project Summary/Abstract We make hundreds of decisions a day. Value-based decision-making requires the orchestration of multiple processes that enable us to learn from prior experience and then use this information to guide our behavior. Deficits in decision-making are common in psychiatric disorders that result from disruptions in how value is estimated or assigned, how value estimates influence action selection, or the ability to make inferences about the environment to guide decisions. These disruptions are unresponsive to current treatments and contribute to the functional disability evident in mental illness. Hence, a deeper understanding of the mechanisms underlying adaptive and maladaptive reward learning is required to address this unmet therapeutic need. Here, we will use fiber photometry, optogenetics, and computational modeling in rats performing two translational behavioral tasks to identify the neurophysiological mechanisms underlying value-based decision-making. The orbitofrontal cortex (OFC) and the striatum are essential mediators of reward processing and decision-making, and both the ventromedial and lateral subregions of the OFC (vmOFC and lOFC, respectively) project glutamatergic neurons to the striatum in a topographic manner; the vmOFC mostly innervates the medial striatum (mS) whereas lOFC preferentially targets central striatal regions (cS). Identifying how these distinct orbito-striatal pathways contribute to specific aspects of value-based decision-making is essential. We aim to (1) identify how dynamic changes in vmOFC→mS and lOFC→cS circuit activity mediate flexible reward learning and (2) determine how alterations in circuit activity disrupt this process. Specific Aim 1 will use dual-color fiber photometry to measure the activity of the vmOFC→mS or lOFC→cS circuits with simultaneous measurement of local OFC parvalbumin-positive (PV+) GABA interneuron activity. Specific Aim 2 will use complementary gain- or loss-of-function optogenetic interventions to confirm the functional relevance of neural activity during behavior. These optogenetic manipulations – targeting orbito-striatal glutamate circuits or OFC PV+ interneurons – will be delivered to (1) enhance the dynamic changes in neural activity associated with optimal task performance or (2) perturb normal neural activity and induce behaviorally distinct disruptions value-based decision-making. Each Specific Aim will evaluate value-based decision-making in rats tested in a probabilistic reversal learning (PRL) or 2-step reinforcement learning task. Computational models of reinforcement learning will provide an in-depth analysis of behavioral performance, and regression analysis will determine how changes in neural activity contribute to task performance. With a multidisciplinary approach and high cell-type- and circuit-specificity, our findings will elucidate the neurobiological mechanisms underlying decision-making and provide critical insight for the development of new and effective therapeutic strategies for mental illness.

Bridgewater

Dissolution of the Village of Bridgewater into the Town of Bridgewater

NINDS - National Institute of Neurological Disorders and Stroke

Project Summary/Abstract Epilepsy affects more than 50 million individuals globally and remains a significant public health challenge, with one-third of patients experiencing drug-resistant seizures. Traditional models focusing on excitation-inhibition imbalance have proven insufficient to explain the full heterogeneity of epilepsy subtypes, highlighting the urgent need for deeper mechanistic insights at the level of neural circuits. To address this need, we propose the 2025 SFN Pre-Meeting on Distinct Circuit Mechanisms in Epilepsies, a two-day scientific conference to be held November 13–14, 2025, at the University of California, San Diego. Aligned with the NINDS mission to advance fundamental knowledge of the brain and nervous system, this meeting will focus on basic neuroscience approaches to understanding circuit-level dysfunction in epilepsy. The conference will include several thematic sessions, including (1) abnormal circuit development, (2) ion channel dynamics in networks, (3) glial contributions to instability, and (4) network mechanisms underlying comorbidities. It will bring together leading investigators across molecular, systems, and computational neuroscience, with presentations spanning in vitro, in vivo, and human models. Trainee poster sessions, structured breakout groups, and mentorship opportunities will further enhance scientific exchange and collaboration. By promoting cross-disciplinary integration and mechanistic rigor, this meeting will serve as a catalyst for novel discoveries in epilepsy research, directly supporting NINDS’s strategic priorities in circuit neuroscience and neurological disorder mechanisms.

NIGMS - National Institute of General Medical Sciences

PROJECT SUMMARY The cellular response to hypoxia is highly conserved yet is strikingly different between mammalian cell types. Even within striated muscle, which represents over 40% of human body mass, the response to hypoxia is distinct between skeletal and cardiac muscles. Our group is interested in understanding how these striated muscle cells respond to hypoxia and how hypoxia-mediated changes in muscle metabolism contribute to muscle-specific outcomes and the integrative response to low oxygen. In the proposed project period, we will investigate 1) the mechanisms of hypoxia sensing and signaling in response to acute and chronic hypoxia; and 2) the integrated systems physiology by which striated muscles interact in settings of low oxygen. We will investigate these mechanisms in three types of striated muscle that we and others have previously demonstrated have distinct responses to hypoxia- skeletal, right ventricular (RV) and left (LV) cardiac muscle. Our approach will develop and utilize new transgenic mouse models, use novel tools to quantify hypoxia, and capitalize on integrative ex vivo approaches to dissect systems-level responses to low oxygen. A multitude of human conditions and environmental stimuli impart changes in local or systemic oxygen levels, yet the specific mechanisms of hypoxia signaling and their integration across systems remains unclear. The proposed integrative studies from multiple muscle cell types will increase our mechanistic understanding of the basic molecular and physiological responses to low oxygen.

NINDS - National Institute of Neurological Disorders and Stroke

PROJECT SUMMARY The ability to seamlessly link individual movements into smooth, efficient sequences is a fundamental aspect of human behavior, enabling everything from speaking to playing musical instruments. This process often feels effortless, yet it relies on complex interactions between different brain regions. When these interactions break down, as seen in movement disorders such as Parkinson’s disease, the ability to learn and execute motor sequences can become severely impaired. The basal ganglia (BG), a set of deep brain structures, is thought to play a key role in this process, but its precise contribution to learning and performance of motor sequences remains unclear. Some prominent theories suggest that the BG store and select well-learned skills, while emerging evidence instead indicates that they act as a tutor, shaping skill learning in the cortex but becoming unnecessary once a skill is well-learned. My project will test these competing ideas by examining how different BG circuits support different stages of motor learning. Using high-density Neuropixel probes, I will record activity from three key cortical regions involved in movement planning and execution—primary motor cortex (M1), supplementary motor area (SMA), and pre-supplementary motor area (pre-SMA)—as well as their anatomically-connected regions in the BG output nucleus, the internal segment of the globus pallidus (GPi). By comparing neural activity during both early learning and well-practiced sequence execution, I will determine how these circuits encode motor sequences over the course of learning at the level of both single-units and neural populations (Aim 1). To test whether these circuits are necessary for learning and recall, I will use pharmacological inactivation with the GABA-A agonist muscimol followed by permanent lesions to disrupt GPi output at different learning stages (Aim 2). This approach will reveal whether the BG is required for learning and performing new sequences but not for executing highly practiced ones, providing insight into how BG circuits interact to shape motor skill acquisition. This fellowship will provide rigorous technical training in multi- area electrophysiology, pharmacological inactivation, and computational analysis of population-level neural data. I will conduct this work within the Systems Neuroscience Institute at the University of Pittsburgh, which offers a uniquely collaborative environment with extensive expertise in nonhuman primate neuroscience. Professional development in mentorship, scientific communication, and leadership will complement my technical training. Together, this research and training plan will prepare me to lead an independent research program focused on the neural basis of skilled behavior.

Distinguished Young Women Foundation

Distinguished Young Women Foundation is an active grantmaking foundation based in Mobile, AL. Focus: general. Contact the foundation directly for current funding opportunities.

NSF

Understanding the structure of massive collections of objects in terms of pairwise distances among their members is fundamentally important to human activities and interests throughout science, industry, and beyond. Here, the notion of “distance” is varied and hence highly expressive: It can include familiar settings such as the 3-dimensional world in which we live and the geometry of curved surfaces, and it covers important notions of proximity within, for examples, social networks, collections of proteins or images, the Internet, and many more; such geometries typically reside in high dimensions and seem nonintuitive or strange in comparison to our Euclidean surroundings. A paradigm that has proven to be ubiquitously powerful and useful to a rich variety of pure and applied investigations is embedding a geometric space of interest into a “nicer” space in which one knows how to perform important tasks (e.g., discovering clusters, efficient storage, compression and optimization, and many more). Typically, this goal cannot be achieved without distorting distances, so the pertinent issue is determining the rate of growth of the smallest possible distortion. This is the overarching theme of the project, which is devoted to resolving central issues and longstanding mysteries of the field. For example, it is not even known how much distances must be distorted, in terms of the total number of objects, when one wishes to represent them in the 3-dimesnional space in which we reside. As another example, while famous and very important work in the 1980s and 1990s determined the largest possible rate of growth of the distortion required to embed a finite metric into Euclidean space (without restricting the target dimension as done above), it remains a longstanding conundrum whether this worst-case scenario can occur when the collection that we wish to embed is a subset of a classical space of vectors (e.g. when distances are measured in terms of the sum of powers of the differences of coordinates, or collections of matrices with natural notions of distance such as the so-called nuclear norm, which occurs in settings ranging from pure mathematics to machine learning and signal processing). When the distortion is shown to grow slowly, as it often does (this entails devising a meaningful geometric embedding), the project harnesses it to devise efficient algorithms for a range of scientific tasks. On the flip side, demonstrating that there is no embedding with small distortion entails obtaining insights and devising tools that are useful and interesting in their own right (e.g., meaningful ways to decompose geometries). The project is devoted to addressing specific open questions in the above directions, as well as deriving unforeseen applications and developing overarching methodologies. The project will include a major effort to train and mentor graduate students and early career researchers through the PI’s advising role and his leadership roles in the mathematics research community. It is common in analysis and geometry that solutions to natural questions about representing objects using other simpler objects cannot be done without incurring large errors. For example, Bourgain’s famous embedding theorem states that any finite metric space can be embedded into a Hilbert space with bi-Lipschitz distortion of order the logarithm of the number of points, and subsequent work showed that this cannot be improved. However, if the given metric space is a subset of a classical space such as a Lebesgue space, a Schatten-von Neumann trace class, or a subset of a discrete group such as the Heisenberg group or the lamplighter group, can one say anything about its embeddability properties other than merely that it is a metric space? It also remains a longstanding open question to determine the growth rate of the largest possible distortion that is required to embed a finite metric space of a given size in 3-dimensional Euclidean space. As a final example, suppose that we are given a 1-Lipschitz mapping from a subset of a finite dimensional normed space to a Banach and we wish to extend it to a Lipschitz function that is defined on the entire ambient space and whose Lipschitz constant is as small as possible: How must this Lipschitz constant grow with the dimension? This is open even for Euclidean space. The project studies the above questions, and many more along those lines. When low distortion is possible, it yields structural insights that are impactful elsewhere, including algorithm design; such applications are pursued in this project after embedding theorems are proved. The project also studies impossibility results showing that large distortion must be incurred; this involves developing new geometric structural insights and tools in harmonic analysis, probability, isoperimetry, etc. This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.

NSF

Advances in sensing, artificial intelligence, computation and communication are enabling the deployment of large networks of autonomous systems, such as drones, ground robots, and other mobile agents, across diverse applications including search and rescue, environmental monitoring, precision agriculture, and transportation. The research funded by this grant aims to create new mathematical tools and algorithms that allow large groups of autonomous agents to operate safely, efficiently, and collaboratively, even under physical constraints, communication limitations, and local decision-making. The central idea of this research is to model large groups of autonomous agents not individually, but as evolving spatial distributions like densities or concentrations over a region. This macroscopic perspective looks to enable the design of scalable, tractable algorithms that guide the collective behavior of many agents, while still accounting for each agent’s physical limitations, local interactions, asynchronous timing, and safety constraints. By linking these global objectives with local decision-making through new optimization techniques, the project seeks to create algorithms that are both theoretically grounded and practically applicable. The research will be complemented by educational and outreach activities that include the development of new curriculum for undergraduate and graduate education, research experiences for students through high-fidelity simulations, and opportunities to engage in algorithm development and visualization at the multiagent robotics (MURO) Lab at the University of California, San Diego. Findings will be disseminated through academic publications, conference sessions, and public engagement efforts. This project aims to develop new tractable, robust, safe and distributed transport algorithms for large-scale multi-agent (autonomous) systems modeled by probability distributions. The theoretical foundation relies on the design of new macroscopic proximal gradient algorithms that account for individual agent limitations as well as approximation errors arising from the use of a finite number of agents. The research is organized around three main thrusts: (i) the development of robust transport algorithms for large-scale homogeneous populations, with robustness characterized via Input-to-State (ISS) stability properties; (ii) the design of safe-proximal gradient algorithms for homogeneous agents incorporating feedback control for macroscopic proximal gradient optimizations; and (iii) the formulation of distributed safe proximal gradients for heterogeneous populations, coordinated by a higher-level network of operators. This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.

NEI - National Eye Institute

Abstract: This project aims to establish distributed federated learning (FL) approaches for multi-task training of foundational machine learning (ML) models for diabetic retinopathy (DR), using multi-modal, real-world optical coherence tomography (OCT) data (OCT cross-section, OCT angiography (OCTA), and OCT enface). DR is one of the leading causes of severe vision loss. Early detection, prompt intervention, and reliable assessment of treatment outcomes are essential to prevent irreversible vision loss from DR. However, there are major challenges towards developing clinically relevant holistic algorithms that can perform multi-tasks, i.e., multi-class classification of disease stages (diagnosis), prediction of onset and progression of disease stages (prognosis), and assessment of treatment outcomes. They require large amounts of well curated and labelled datasets from a diverse sub-population for robust performance. Moreover, efforts towards large, centralized datasets for ML research are hindered by significant barriers to data sharing and privacy concerns. In this project, we propose to develop foundational ML models that allow efficient learning of feature representations from a large corpus of ophthalmic imaging data for various downstream tasks – breaking the task-specific paradigm of current ML models. We also establish novel federated ML approaches, where the model training is distributed across institutions instead of sharing patient data. Our first aim is to establish and validate a domain adaptive FL framework for DR diagnosis across four independent institutions. We propose a novel ophthalmic adaptive personalized FL (optho-APFL) technique to tackle domain shift caused by heterogeneous data distribution at different institutions (due to different sub-population density and OCT devices/imaging protocols). We will conduct experiments on the FL deployment in a clinical setting and integrate a granular differential privacy (DP) algorithm into our FL framework to provide ‘patient-level’ data privacy. Key success criterion is to deploy the FL framework and validate FL-trained ML models against state-of-the-art models for DR diagnosis. The second aim is to develop foundational ML models with self-supervised learning (SSL) to learn multiple tasks within the same framework from label invariant OCT/OCTA data, where different institutions don’t need to have labeled data for each of the tasks. We will train these foundational models in a centralized and FL framework for comparative analysis. Key success criterion is to i) validate foundational model performance for multi-task learning (MTL) (DR staging, prediction of NPDR to PDR progression, and prediction of DME treatment evaluation) on new clinical data (centralized and FL approach, and ii) identify task-specific quantitative OCT/OCTA (mean and artery-vein specific) features. As an alternative approach, we propose diffusion probabilistic modeling (DPM) for SSL to learn holistic representations from multi-modal OCT data for MTL, and to explore dynamic federated averaging approaches. Success of this project will establish OCT/OCTA based distributed foundational models for objective MTL in DR using label-invariant data across multi-institutions and standardize OCT/OCTA features for MTL.

NSF

In today's rapidly evolving intelligent systems landscape, spanning various smart operations such as traffic and energy grid management to deploying sensors for environmental monitoring, optimizing resource allocation is crucial for efficient decision-making and policy formulation. Despite numerous advances in the field of optimization theory, current methods for optimal resource allocation problems often falter, especially in large, distributed networks where devices must coordinate without the presence of a central authority, resulting in slow solutions and high communication demands. This project addresses these fundamental limitations in combinatorial optimization problems, specifically submodular maximization under matroid constraints—an NP-hard problem that regularly appears in optimal discrete resource allocation. While efficient mathematical approaches have been developed to achieve near-optimal solutions in centralized form for submodular maximization under matroid constraints, these solutions often incur significant computational costs. In distributed settings, in addition to these computational costs, the optimal resource allocation process is further exacerbated by high in-network communication costs and local data disclosure issues. Our objective is to develop practical solutions that reduce in-network communication costs, maintain favorable trade-offs in optimality gaps, and accelerate solutions. The project will deliver practical distributed algorithms with analytically characterized optimality gaps and proven practicality measures. The project will also provide vital educational and training opportunities for students from K-12 to graduate level, preparing the future workforce in optimal decision-making for autonomous systems. This research aims to significantly advance the theory and practice of distributed submodular maximization under partition matroid constraints. We will focus on the continuous relaxation approach, which has improved the optimality gap for centralized solutions from 0.5 to 0.63. Our primary goal is to develop practical distributed algorithms based on this approach that maintain strong optimality guarantees while drastically reducing communication and computational costs. We will achieve this through two key technical avenues: first, by exploring methods like hard-thresholding and localized gradient approximations to limit data exchange among agents; and second, by accelerating convergence rates using novel variance reduction techniques based on control variates and advanced algorithms that reuse gradient information. The project emphasizes rigorous formal analysis to characterize optimality gaps and will validate these algorithms through extensive computer simulations and experimental studies, including applications in multi-agent robotic coverage and smart grid systems. Outcomes will include a catalog of provably correct distributed optimization strategies and open-source software tools. This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.

Do-It-Best Corp

Distribution Center Expansion

Lusted Water District

Distribution System Improvements

City of Joseph

Distribution System Improvements 2022

Lakewood Homeowners Inc.

Distribution System Replacement Feasibility Study

Christmas Valley Domestic Water Supply District

Distribution System Replacement Phase 3