New York University

NSF Awards · FY 2025 · 2025-10

Despite recent technological advancements and shifts in work patterns, traffic congestion continues to burden U.S. cities, resulting in the loss of millions of dollars due to wasted fuel and lost productivity. In congested urban areas, these issues are exacerbated by the time and stress associated with searching for parking. The standard strategy toward travel-demand management to tackle traffic congestion is road and parking pricing, but the real-world implementation of pricing approaches has proven to be difficult due to concerns about their fairness. This project proposes an alternative in the form of integrated parking-management and ridesharing systems that reduce traffic congestion through non-pricing mechanisms. In addition to alleviating congestion, the proposed systems aim to expand access to opportunities through collaborative ridesharing solutions that are deployable even in auto-dependent communities with limited transit options. This project aims to develop a data-driven travel-demand management framework for the deployment of an integrated parking-management and ridesharing cyber-physical system (CPS). The CPS will enable the implementation of integrated parking-management and ridesharing schemes to mitigate congestion and enhance mobility by jointly coordinating parking access and collaborative peer-to-peer carpooling programs. The proposed framework advances research on peer-to-peer ridesharing and travel-demand management through optimization models that jointly consider hybrid travel-incentive structures, measures of personal affinity between travelers, and user-specific valuations of these measures in the ride- and parking-matching process. The framework envisions a CPS architecture that dynamically updates users’ affinity measures and accounts for trade-offs they make between travel costs and trust considerations. Beyond new policies, models, and heuristics, the innovation of this project lies in the proposed CPS, which combines traveler digital twins, machine-learning-based predictions of parking-facility states, and internet-of-things technologies to operationalize the proposed travel-demand management schemes. Lastly, the project will establish a living laboratory to evaluate the performance of the integrated parking and ridesharing CPS under real-world conditions. 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 Awards · FY 2025 · 2025-10

Machine learning (ML) and artificial intelligence (AI) technologies are being adopted for a wide range of applications. This has spurred interest in the use of ML and AI for chip design. Chip design is currently heavily automated using electronic design automation (EDA) software. Recent work has shown that AI and ML methods can further increase the level of automation as well as improve the quality of existing EDA tools. However, ML methods pose their own dangers and risks. They have been shown to be easily tricked by small changes in their inputs. They can also be easily "backdoored" by modifying only a tiny fraction of their training data. While these risks have been extensively studied in other domains, their impact has not been extensively examined in AI/ML based EDA and chip design tools. This project's novelties are (1) the first comprehensive look at the impact of input and training data perturbations and attacks on the quality, performance and security of AI/ML based EDA tools; and (2) the first thorough investigation into mechanisms to defend against such attacks. The project's broader significance and importance is that enables the trustworthy adoption of AI/ML methods in the chip design industry, resulting in greatly enhanced productivity and chip design quality, while ensuring trustworthiness. The project pursues these aims in three research thrusts. Thrust 1 focuses on discovering meaningful and contextual perturbations of inputs to the different steps in the EDA flow, starting from design specification to logic synthesis, test-point insertion and physical design. To this end, the project investigates a new "EDA vs. EDA" threat model, where tools from competing vendors in the same seek to degrade each other's performance by injecting targeted functionality-preserving transformations in the inputs of a downstream tool. Thrust 2 evaluates the impact of training data poisoning and backdooring attacks on ML-based EDA tools spanning both pre- and post-silicon use cases. This Thrust demonstrates how carefully inserted stealthy triggers like netlist and layout patterns, comments in RTL code or temporal sequences of instructions can result in undesirable outcomes. Thrust 3 builds robust ML-based EDA tools that can withstand attacks demonstrated in the prior two thrusts. This Thrust explores techniques for meaningfully infusing ML for EDA with constraints from the relevant EDA domains during model training. Overall, the three Thrusts synergistically work together to create the foundations for next-generation trustworthy AI/ML-native EDA, while also training hardware design students with a fundamental understanding of security and trust concerns in AI/ML. 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 Awards · FY 2025 · 2025-10

The exploration of Terahertz (THz) communications (above 100 GHz) is promising for achieving ultra-high data rates and bandwidth in next-generation wireless networks. The combination of large transmitting arrays and smaller wavelength in THz communications introduces new challenges where the near-field range of a transmitter can extend to several meters. A wireless base station or access point may serve multiple near-field and far-field users. This project investigates the impacts of near-field complexities and the interplay between near- and far-field communications. It leverages near-field properties, such as the generation of curved beams for going around the obstruction or reconstructing the beams after interacting with an obstacle, to improve blockage resilience and system performance of THz wireless networks. This project also includes significant efforts to motivate participation in STEM, at all educational levels. This project aims to explore near-field propagation properties of wireless signals to enable blockage-resilient high-speed multi-user wireless communications above 100 GHz. The proposed research comprises four thrusts, (1) understanding and modeling the fundamental limits of near-field wavefront shaping for realizing agile and high-speed communications in the scenarios with link blockage, (2) investigating the design and fabrication of a new programmable array architecture that can generate near-field wavefronts and dynamically adapt them according to the mobility and signal transmission blockage conditions, (3) designing and implementing digital signal processing algorithms to support 10s of GHz of bandwidth in a power-efficient manner, and (4) developing foundations for interference modeling and multi-user communications with the consideration of both near- and far-field regions. Through the physics of wave propagation, mathematical modeling, hardware design, and experimental demonstration, this project will lay out the foundation to bridge the theory of near-field electromagnetics with practical wireless networks. 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 Awards · FY 2025 · 2025-10

An emerging form of data processing named cryptographic computing enables computation on encrypted data, providing unprecedented levels of privacy and security. However, these techniques are not widely used today as they run very slowly on existing processors and hardware. This project will implement the infrastructure needed to enable researchers to develop new hardware designs that can overcome these performance limitations and enable cryptographic computing. Throughout this project, named Cryptolets, the team will implement the hardware and software needed to accelerate core cryptographic computing functions in hardware, the infrastructure to connect these pieces, and software to validate it. Cryptolets will focus on three intertwined efforts. First, it will implement and open source a hardware library of key operators and kernels used in cryptographic computing (e.g., modular multiplication, number theoretic transform, and multi-scalar multiplication), and prototypes of existing full protocol accelerators (e.g., homomorphic encryption). Second, it will develop infrastructure needed to connect and build chips. This includes automated scripts to run necessary hardware tools and chiplet interfaces. Since cryptographic computing kernels are complex and large, Cryptolets provides users the ability to break designs into smaller chips and connect them. Third, it will provide tooling support for formal verification, testing for side-channels, and test vectors for validation. Cryptolets will develop suites of educational resources and conduct numerous community building events to advance cryptographic computing. First, the Cryptolets team will develop, record, and release lectures on the fundamentals of cryptographic computing, hardware for cryptographic computing, chiplets, formal verification, and benchmarking. Community building exercises will be done via workshops and online events for experts and newcomers across disciplines to connect. Finally, by releasing Cryptolets infrastructure, the barrier to entry is significantly lowered. Coupled with education and community building activities, the research community is expected to grow markedly. All aspects of the project (e.g., code, lectures, slides, and papers) will be made publicly available for all to access and use. All Cryptolet material and code can be found on the website: cryptolets.org. Websites and repositories are expected to be hosted indefinitely. In addition, archival records of the Cryptolets project (e.g., code, data, reports) will be taken and uploaded at the midway (end of year 2) and final (end of year 4) phases of the project. This will provide a permanent and open backup on the internet for all to access. 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 Awards · FY 2025 · 2025-10

Accurate interpretation of hyperspectral data depends on the availability of reference spectra: measurements of known materials compiled into spectral libraries. Such libraries support both direct classification and machine learning applications. When combined with on-site hyperspectral imaging, they have proven effective across a variety of domains including heritage conservation, homeland security, hydrology, and geology. Urban conditions, however, present unique challenges to spectral data collection. In this context, urban materials refer to the components of the built condition, including both manufactured materials (e.g., asphalt, concrete, paint) and naturally occurring materials that have been anthropogenically modified for urban use (e.g., cut stone). Although hyperspectral data have been utilized in select urban planning tasks, the broader potential of hyperspectral imaging for material identification remains underutilized. This project will develop and extend community involvement in hyperspectral remote sensing technology to analyze and study urban landscapes. This will be paired with open metadata standards, modular processing toolkits, and automated archival workflows that prioritize FAIR principles. HS-SPECTRA (Hyperspectral Standardizing and Sharing Possibilities for Urban Conditions through Toolkits, Resources, and Archiving) addresses fundamental challenges in hyperspectral library design by: 1) Developing a metadata architecture tailored to longitudinal urban field campaigns; 2) Incorporating auxiliary sensors to contextualize spectral variability; 3) Implementing a flexible versioning and querying model that reflects the dynamic nature of repeated, real-world observations; and 4) Enabling interoperability across platforms through spectral resampling and standardization pipelines. By embracing the temporal complexity of real urban conditions and focusing on reproducible, extensible data infrastructures, HS-SPECTRA will generate a uniquely valuable dataset for cross-instrument, cross-temporal analysis. The resulting protocols and open-source tools will significantly advance methodological rigor and accessibility in urban planning, Earth system science, computer vision, and urban monitoring fields. This award by the Office of Advanced Cyberinfrastructure within the Directorate for Computer and Information Science and Engineering is jointly supported by the Directorate for Engineering. 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 Awards · FY 2025 · 2025-10

1927468 (Grimm) 1927167 (McPhearson). Cities and urbanized regions worldwide are exposed to extreme weather events and rising seas. They are at risk because their infrastructure often is in disrepair, no longer appropriate for more intense or frequent extreme events, or unable to keep up with rapid urban population growth. Traditional engineered infrastructure, such as stormwater drainage systems or sea walls, is usually designed for only one purpose and seldom can adapt to changing conditions. Solutions that are based on nature-preserving protective ecosystems, incorporating ecological elements or even mimicking nature in built infrastructure, offer flexibility in the face of changing conditions and provide multiple benefits to society, often at relatively low cost. The NATure-based solutions to Urban Resilience in the Anthropocene (NATURA) project links 26 networks to enhance connectivity among the world's scholars and practitioners and improve the prospects for global urban sustainability. NATURA exchanges knowledge, shares data, and enhances communication among research disciplines and across the research-practice divide to advance understanding of how to prepare for the growing threat of extreme weather events. As an important part of this knowledge sharing, learning exchanges will build capacity of the next generation of researchers and practitioners to work together on applications of nature-based solutions in a wide range of social, ecological, and technological contexts. The NATURA international network of networks brings together research scientists (ecologists, engineers, and social scientists) and city practitioners (such as officials from city, county, or state governments, members of non-governmental organizations, and community leaders) who work on devising and implementing solutions to the challenge of extreme events. NATURA will unite 21 networks focused in Europe, South Africa, China, North and South America, and globally with 5 U.S.-based networks that are conducting research and implementing nature-based solutions. NATURA will advance theory and research on nature-based solutions, identifying and filling research gaps across diverse global social-biophysical contexts to understand where nature-based solutions are unique or can be more generally applied to meet urban resilience challenges. Through all-hands meetings, thematic working groups, regional nodes, and synthesis writing workshops, the project will accomplish the goals of synthesis and data sharing, and network coordination. Early-career researchers and practitioners will be sponsored by NATURA to pay five-week visits to network partners. Further, NATURA will train postdoctoral scholars and graduate students through learning exchanges to networks around the globe. Through collaboration with partners, international students will be invited to participate in these exchanges, hosted by US networks. The Accelerating Research through International Network-to-Network Collaborations (AccelNet) program is designed to accelerate the process of scientific discovery and prepare the next generation of U.S. researchers for multiteam international collaborations. The AccelNet program supports strategic linkages among U.S. research networks and complementary networks abroad that will leverage research and educational resources to tackle grand scientific challenges that require significant coordinated international efforts. This project was co-funded by the Division of Environmental Biology (BIO/DEB). 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 Awards · FY 2025 · 2025-10

Software bugs can have disastrous consequences, ranging from financial costs to loss of human life. As a result, for high-stakes systems, software vendors are increasingly applying techniques that can prove the absence of various kinds of bugs. However, existing techniques have limitations that make them inapplicable for certain types of programs that make use of randomness, which is common in sensitive software domains such as cryptography and machine learning. This project will develop new techniques for reasoning about randomness in programs, which will make it possible to prove important properties about these programs, thereby improving software quality in these critical areas. In addition, the team of researchers will develop educational materials to make the project's ideas more broadly accessible to students, researchers, and industrial practitioners. This project targets programs that exhibit two important kinds of effectful features: concurrency and randomization. Existing formal verification techniques cannot handle the complexity and expressivity of many programming language features, and these features make it harder to write, test, and reason about programs. Establishing correctness in the presence of just one of these features is hard enough, and it only becomes more difficult when they are combined. This project will develop program logics and reasoning tools that can enable more precise, compositional analysis of concurrent randomized programs by building on a new semantic model of randomness and concurrency. The investigators will formally verify the soundness of the logic and build a framework for using it inside of an interactive theorem prover. This formalized framework will facilitate further breakthroughs in verification of concurrent randomized programs in different domains. 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 Awards · FY 2025 · 2025-10

Systems based on modern large language models (LLMs) play an increasing role in how users access information and compose text. For instance, a user executing a web search will increasingly rely on LLM-based systems to summarize their search results, rather than viewing individual web pages, and they might use LLM-based systems to “talk to” long documents like financial reports, rather than reading them in their entirety. To support these new paradigms, it is important that an LLM be able to generate responses that are factual, informative and safe. However, satisfying these criteria is not sufficient: a response should also be at the right level of abstraction or detail, in the right format, creative where appropriate, and aligned with other user needs. Current practice has neglected evaluation of these more subtle factors. This project proposes to address these shortcomings by identifying a set of “evaluation concepts” to indicate the kinds of areas where LLMs are failing, like “lack of detail in a list.” The project will then develop technology for automatically evaluating and improving LLM responses according to these concepts. This project aims to improve the evaluation and the functionality of LLMs in two ways. First, the project will discover a concept taxonomy and learn how to evaluate LLM responses according to the concepts in that taxonomy. This process will necessitate advances in reward models, which are themselves LLMs, customized to reliably score responses. Second, these reward models are applied to actually improve the LLMs’ responses. Specifically, the project will curate training data exhibiting the correct kinds of behavior for each concept, enabling training of LLMs that do better on those concepts. Finally, the project will develop methods for iteratively improving responses using our reward models. The project will open-source the concept taxonomy and reward models that will outperform closed-source, proprietary models. These models will enable the public to have a better sense of the performance of LLM systems across a variety of applications, and will drive the open-source community to build stronger, more reliable LLM systems. 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 Awards · FY 2025 · 2025-10

This award is funded in whole or in part under the American Rescue Plan Act of 2021 (Public Law 117-2). Modern question answering systems, embedded in search engines and digital assistants, have improved dramatically with the development of large neural network models. When a user asks a simple question, these systems can typically return an answer directly rather than just linking to a webpage. However, these systems still fail on more complex questions, and when they fail, they may mislead their users. They lack an important capability that humans have: the ability to reason about and synthesize the information they see, retrieve and integrate additional information, and arrive at a justified conclusion. This CAREER project aims to address this shortcoming by developing systems that "think through" textual evidence, leading to more reliable answers that can be explained to a user. Such advances fit into a broader thread of building trustable artificial intelligence (AI) systems that explicitly show their work and are auditable before and during their deployment. This project specifically addresses the problems of question answering and fact-checking by developing a learning-based system that reasons in natural language. The system takes text as input, then applies pre-trained neural network models to reformulate that text, derive conclusions from it, and eventually check a claim or verify an answer. This process produces a series of logically connected statements understandable by a human. This outcome is enabled by two modules. First, a deduction module repeatedly combines two statements and generates a third that follows from the inputs, encapsulating common logical rules. Second, a verifier determines whether the final deduced evidence validates the original claim. Both systems are built from pre-trained models like T5 that have demonstrated strong generalization capabilities. Collecting training data for these models constitutes a core challenge; the project's approach blends multiple strategies including synthetic data generation and human-in-the-loop annotation. These techniques are applied to the domains of question answering and fact checking, problems where providing additional explanation and justification instead of just giving a best-effort answer are essential to make usable systems. This system paves the way for NLP tools that know what they don't know, provide interpretability for end users, and enable system developers to better understand and improve their models. 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 Awards · FY 2025 · 2025-10

This project will help us understand how galaxies, like our Milky Way, grow and change over time. Galaxies are filled with gas that forms stars and fuels black holes. When stars explode as supernovae, they send out large amounts of energy and material into space, creating galactic winds that can shape how galaxies develop. These winds are important because they can stop galaxies from becoming too massive and full of gas. However, there is a big difference between what leading computer models predict about these winds and what we actually observe. The investigator will use advanced computer simulations to study how these winds interact with the gas that surrounds galaxies, known as the circumgalactic medium (CGM). By combining these simulations with real observations, the project will give us new insights into how galaxies evolve. This work will involve students, helping train the next generation of scientists and improving our understanding of the universe. The investigator team will address the glaring discrepancy between current cosmological galaxy formation models and observations of galactic wind outflow rates by focusing on the interaction between galactic winds and the CGM. Using the state-of-the-art adaptive mesh refinement (AMR) magnetohydrodynamics code framework, athena++, the investigators will conduct high-resolution simulations that capture the self-consistent launching and dynamics of these galactic winds and their interactions with the CGM. These simulations will include key physical processes such as magnetic fields, thermal conduction, and cosmic rays, UV shielding, thermal conduction, and non-equilibrium ionization, assessing their individual and collective impact. Crucially, the use of AMR in these simulations will enable unparalleled spatial fidelity, from the interiors of galaxies to the expansive CGM. A robust forward modeling pipeline will be developed to produce multi-wavelength mock observations, enabling direct comparisons with observational data from instruments such as the James Webb Space Telescope (JWST). This collaborative team includes with experts in simulations, theory, and observations. 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.

NIH Research Projects · FY 2025 · 2025-09

PROJECT SUMMARY / ABSTRACT. Research Project: Spatial memory – memory of where an event happened or an object was located – depends on the hippocampus in a wide range of vertebrate species, including mammals and birds. In humans, most spatial memories are formed through visual experience. However, it is unclear how visual information is processed by hippocampal memory circuits to support spatial memory formation. Two obstacles have hindered answering this question: 1) the complexity of networks in which the hippocampus is embedded in mammals, and 2) the need to observe many independent memories being formed and recalled at identifiable moments in time. This project overcomes these obstacles by leveraging the unique advantages of the black-capped chickadee Poecile atricapillus, a food-caching bird that depends on an intact hippocampus to retrieve previously hidden food items. Chickadees rely predominantly on vision for navigation, as in humans. They also form many independent memories at precisely identifiable times. Finally, the neural pathways carrying visual input from the retina to the hippocampus in birds are relatively simple. I will exploit these features to dissect the transformation of lower-order visual inputs into the observed spatial firing patterns and memory functions of the hippocampus. I will (1) dissociate visual and spatial representations in the chickadee hippocampus using a novel discrete foraging task, (2) compare the visual and spatial representations in a visual cortex analog that provides monosynaptic input to the hippocampus, and (3) causally test the role of this visual input pathway for hippocampal coding and memory. Hippocampal circuits are highly similar in mammals and birds, so the results promise to reveal fundamental computations that are shared across vertebrates. Further, this project is broadly relevant for hippocampal disorders including Alzheimer’s disease, which is associated with visuospatial deficits and altered eye movements. Candidate and Career Goals: I aim to establish an independent lab studying how visual information is processed by memory circuits. I have a background in engineering and vision-related systems neuroscience, and have made foundational discoveries in the hippocampus of food-caching birds, including recently publishing the first neural recordings in these species. This history, combined with the scientific and professional training planned during the K99 phase of this project, positions me uniquely to succeed in my goals. Career Development Plan: I will be trained by mentors Dr. Dmitriy Aronov and Dr. Larry Abbott at Columbia University, and Dr. Elizabeth Buffalo at the University of Washington. Dr. Aronov is an expert in the experimental techniques I will learn, while Dr. Abbott is a world-renowned theoretical neuroscientist who will provide training in the analysis and modelling of complex datasets. Dr. Buffalo pioneered the study of visual representations in the primate hippocampus and will advise on task design and analysis, while ensuring that the project is broadly relevant to hippocampus researchers across species. All mentors will provide career development training and advice for my transition to independence.

NIH Research Projects · FY 2025 · 2025-09

The purpose of this proposed study is twofold: First, we propose to develop a deeper understanding of the experiences of severe maternal morbidity among survivors, their families, and their communities, and explore the boundaries of what is considered severe maternal morbidity (SMM) versus other severe complications of pregnancy. Second, we propose to develop community-driven intervention strategies to improve maternal, child, and reproductive health outcomes. All work will be guided by and developed in collaboration with an established community advisory board focused on SMM and maternal health. In Aim 1 we will use a lifecourse lens to conduct in-depth interviews and surveys with survivors of SMM. For Aim 2 we will recruit partners, family members, and support persons of SMM survivors to participate in interviews and surveys to understand their unique experiences which have often been overlooked. In these aims we will explore and describe the experiences, challenges, assets, and strengths of women affected by severe maternal morbidity and other serious complications of pregnancy; and characterize the experiences of partners, family members, and support persons. We will identify long-term consequences of SMM, evaluate social drivers of health of SMM survivors, and better define the boundaries of what is considered SMM from the patient and community perspective vs. a medical or epidemiologic perspective. We will identify causes, contributing factors, and consequences of SMM and other harms experienced by survivors and their families or support persons; survivor, family, and community strengths; health systems structures; and policy implications. In aim 3 we will be guided by causal mapping from Aims 1&2 and community-based prioritization of these factors in selecting priorities for intervention. We will then prototype, in collaboration with SMM survivors and family members, community partners, clinicians, community birth workers, administrators, and policy makers, an intervention strategy or strategies to address their priority concerns in maternal, child, and reproductive health. We will use the 6 Steps to Quality Intervention Design model to enhance prototype readiness for feasibility and acceptability testing as well as future large-scale testing, implementation, and dissemination. This study integrates the NINR strategic plan lenses prevention and health promotion, and systems and models of care to develop community-driven interventions to prevent SMM; mitigate the impact of SMM on survivors and their communities; and promote positive maternal, child, and reproductive health outcomes.

NIH Research Projects · FY 2025 · 2025-09

Project Summary/Abstract Cervical cancer burden falls heavily on women in low- and middle-income countries (LMICs). Cancer control policies have the potential to address this burden by enhancing the reach of evidence-based primary, secondary and tertiary clinical services, and the fidelity with which they are delivered—however, these policies must be implemented if they are to achieve population health improvements. However, we lack a robust, empirically- grounded, theoretically-driven understanding of the determinants of policy implementation success, nor of strategies that could improve policy implementation. This project engages a network of experts from African LMICs and global agencies through: (A) a three-round Delphi process to identify determinants of policy implementation success in LMICs, beginning with the CFIR (version 2 plus modifications for LMICs); (B) qualitative interviews, leveraging the CFIR and Bullock et al.’s policy implementation framework, to understand the process and mechanisms by which these determinants operate; and (C) another three-round Delphi process to identify strategies, from a subset of those in the ERIC compendium, that could be used to strengthen cervical cancer policy implementation in African LMICs. This study focuses on cervical cancer policy, but has relevance for other cancer policies and other public health policies more broadly. The findings of this research can equip researchers and policymakers with empirically- informed, expert-selected strategies that could be implemented and evaluated. To our knowledge, this is the first study to systematically engage global experts in a theory-based, rigorous process to identify both determinants of policy implementation success and strategies to address these determinants. The project will leverage expertise of the MPIs in global cancer control and policy-focused implementation science, a Scientific Advisory Board that includes implementation science experts and global cancer control experts, and an Expert Panel of ~60 policymakers and other experts from African LMICs. There has been much emphasis on crafting and introducing cancer control policies and plans in countries around the world. We need to improve our understanding of how these are being implemented, and how this implementation could be improved so that greater population health improvements can be achieved. This research will generate the necessary building blocks—grounded in D&I science and modified for the LMIC context through rigorous expert engagement processes—to strengthen policy implementation, and ultimately work toward cervical cancer elimination globally.

NIH Research Projects · FY 2025 · 2025-09

The social determinants of health (SDOH) are critical for understanding the intersections of the multitude of individual, interpersonal, community, and societal-level factors that influence a person’s health. Understanding the SDOH in the context of the biological, behavioral, physical/built environment, sociocultural environment, and health system domains of influence is another critical skill to build. New Americans (NA) are persons who have come to live and work in the United States (US) and comprise 15% of the population, 77% of whom are naturalized citizens or lawfully present. They have two unique SDOH factors that will further influence and add complexity to understanding associated health outcomes: Nativity and the migration experience. Nativity—meaning the country of origin of a person who has migrated—is a recognized SDOH. The migration experience is not included as an SDOH, yet it will strongly influence a person’s health outcomes and those of their family members as the nature of voluntary vs. involuntary migration produces differential impacts on health. Many do not speak English well enough to communicate safely in health care settings, which will also affect their health outcomes. Overall, the SDOH specific to NAs are understudied and yet critically important for ensuring healthy populations across the lifespan. There is critical need to build research capacity specific to the social determinants of the health of NAs amongst nurse researchers. Through an interdisciplinary lens, the proposed 3-year R25 will train up to 30 early-career nurse scientists (<5 years into a research-intensive position) to study the SDOH within the context of NA populations. The hybrid 13-week program (1 week in-person, 12 weeks virtual) will help trainees to understand the mechanisms across levels and domains of influence that lead to variations in the SDOH and health outcomes of NAs when compared to their US-born counterparts and develop solutions to address them. Experts from the National Immigration Law Center and the Migration Policy Institute will contribute. To ensure the long-term success of participants, sustained mentoring beyond program completion will occur throughout the 3-year program and develop the mentoring skills of alumni participants. Program graduates will also form a new national network of nurse researchers who specialize in research with NA populations. They will help expand national capacity for research on the SDOH of NA populations as well as nurse researchers with those skills. Materials developed from the program will also be made publicly available for use with other nursing and health professions educational programs. The program’s goals help achieve all facets of the strategic plan for the National Institute for Nursing Research.

NIH Research Projects · FY 2025 · 2025-09

We propose evaluating an adaptive dietary intervention leveraging continuous glucose monitoring (CGM) and the translated and adapted behavioral intervention in Asian Americans with type 2 diabetes (T2D). Despite being a fast-growing group in the U.S., Asian Americans face prominent diabetes care disparities. Compared with non-Hispanic Whites, Asian Americans have a higher prevalence of T2D, a higher incidence of T2D with lower body mass index, worse glycemic control, and poorer health status. They also have the highest prevalence of undiagnosed diabetes among all racial/ethnic groups. Dietary management is key to optimal glycemic control. However, limited dietary interventions have been tailored to Asian Americans’ unique barriers. These barriers include maintaining traditional Asian dietary patterns such as a high intake of starches (e.g., white rice, noodles) that significantly contribute to postprandial hyperglycemia and high glucose variability, consuming foods with high sugar and fat because of acculturation, lack of linguistically translated healthcare programs, unfamiliarity with the U.S. healthcare system, and rapid increase in income discrepancy. Our adaptive dietary intervention, which integrates CGM and translated and adapted Glycemic Excursion Minimization (GEM), targets glucose excursions rather than weight loss. This approach offers a feasible, paradigm-changing solution for Asian Americans with T2D, as over 80% of Asian Americans with T2D are not obese, and conventional lifestyle interventions targeting weight loss are not applicable. Our objective is to evaluate an adaptive dietary intervention leveraging CGM and translated and adapted GEM for Asian Americans with T2D. We will examine the feasibility and acceptability (Aim1) and effect (Aims 2&3) of the adaptive dietary intervention over 24 weeks. We propose to enroll 120 participants (60 Chinese Americans and 60 Vietnamese Americans) with T2D, who will be 2:1 randomized to one of two arms: adaptive dietary intervention and standard of care (SC). Recognizing high levels of heterogeneity in response to the same intervention, based on the principle of adaptive intervention design, our 12-week intervention will start with CGM use only during weeks 0-4. At week 4, participants who achieve the glycemic control goal (10% increase in time in range from baseline) will continue with the CGM alone during weeks 4-12 (CGM alone); otherwise, the translated and adapted GEM will be augmented (CGM-GEM). Our central hypothesis is that CGM alone will be better than SC and CGM-GEM will be superior to CGM alone to improve glycemic control, quality of life, and diabetes distress for Chinese and Vietnamese Americans with T2D. This R01 responds directly to PAS-23-086 (Small R01s), aiming to provide preliminary data and lay the foundation for future large-scale studies. We seek to shift the current dietary management paradigm for Asian Americans with T2D to reduce health disparities in diabetes management.

NIH Research Projects · FY 2025 · 2025-09

Only about two-thirds of people who have mental health (MH) concerns initiate care after a referral is made, and less than a third attend enough sessions to see minimally adequate benefits. For mothers in the child welfare (CW) system living with serious MH conditions, this gap is particularly pronounced. Diminished access to MH services places CW-involved mothers at increased risk for poor functioning and their children at increased risk for child maltreatment. Improving access for CW-involved mothers improves their well-being and reduces risk for their children, thus, the proposed project is of great public health significance. The present study asserts that lack of access occurs for two key reasons – first, the majority of CW-involved mothers who need MH services are not referred for care and, second, even when they are referred, few actually begin services. In CW contexts, referral by a caseworker sets off a chain of events that should end in mothers’ initial use of MH services (hereafter known as the referral-access pipeline). The present application explores determinants of caseworker-driven referral as well as factors that impact the post-referral gap (i.e., the discontinuity in the referral access pipeline between referral and initial use of services). The proposed study is embedded within an ongoing two-arm randomized controlled trial (R01HD102528) that is testing the delivery of a MH intervention, Parenting-STAIR (PSTAIR), within 10 sites providing CW preventive services in New York City. The parent study offers an ideal venue within which mothers’ access to care can be studied, as the gulf between referral and initial use of MH services (the boundaries of the referral-access pipeline) is well-documented in this context. The referral-access pipeline will be explored from two vantage points: the perspectives of caseworkers (N=30) at each of the 10 preventive agency sites and the perspectives of mothers (N=32). The proposed study will utilize convergent mixed methods, including primary qualitative interview data and secondary quantitative data, to identify the individual and contextual factors that facilitate or impede progress along the referral-access pipeline for mothers in the context of the CW system. Coincidence analysis will be used for analyses and is a configurational comparative method that can be used for purposively sampled, small sample sizes. Data from primary semi- structured interviews and secondary survey data from the parent study will be analyzed. The outcomes in the proposed study are referral and initial use. The CNA approach integrates qualitative and quantitative data to understand how combinations of one or more factors may be needed for an outcome (here, referral rate and initial use) across cases. The proposed study answers a call put forward by the National Institute of Mental Health to engage stakeholders to better understand persistent issues related to accessing MH services. The applicant will complete advanced methods training, develop connections with MH services researchers, and build on a research agenda focused on transforming access to quality MH services for mothers affected by mental illnesses and their children who are at risk, promoting optimal outcomes across generations.

NIH Research Projects · FY 2025 · 2025-09

PROJECT SUMMARY/ABSTRACT Visceral pain is a debilitating symptom of irritable bowel syndrome (IBS), a highly prevalent disorder that impacts >11% of people worldwide. Visceral pain results from increased excitation of extrinsic primary afferent neurons (ExPANs) that project from the intestinal tract to the spinal cord. Clinical and experimental evidence suggest that IBS-related pain is caused, at least in part, by dysregulation of serotonin (5-HT), a key regulator of ExPAN sensitivity; however, current strategies to modify 5-HT signaling as a pain therapy have limited efficacy and significant adverse effects. To determine how 5-HT can be better targeted for visceral pain treatment, the proposed studies focus on the impact of gut epithelium-derived 5-HT on ExPAN function. Enterochromaffin (EC) cells of the epithelium produce most of the 5-HT in the gut, which stimulates ExPANs to promote sensory signaling. The serotonin reuptake transporter (SERT), present throughout the gut epithelium, rapidly inactivates 5-HT. My preliminary data strongly suggest that epithelial-restricted 5-HT modulation is a promising target for pain therapy that would limit adverse effects. With the support of a K01 Mentored Research Scientist Development Award, this proposal will define how epithelial 5-HT impacts ExPAN signaling to produce pro- or anti-nociceptive effects. I will accomplish this in two specific aims, with mentored training in electrophysiology, single cell molecular analysis, and biophysical assays to assess 5-HT receptor trafficking. Aim 1: How does gut epithelial 5-HT regulate ExPAN activation? I will determine how availability of epithelial 5-HT impacts ExPAN mechanosensitivity and excitability, which are neuronal properties that directly influence pain perception. Aim 2: Does gut epithelial 5-HT regulate expression and function of ExPAN 5-HT receptors? I will determine how increased epithelial 5-HT affects 5-HTR signaling and trafficking in ExPANs, which will have the potential to reveal intracellular targets for visceral pain treatment. My mentoring team and I have designed this proposal to provide me with the necessary research training and professional development to establish an independent academic career in the fields of neurogastroenterology and pain signaling.

NIH Research Projects · FY 2025 · 2025-09

Dental caries is the world’s most prevalent noncommunicable disease, with considerable health disparities found in low-income rural white children and low-income urban Hispanic/Latino and black children. To combat this silent epidemic, school-based caries prevention can increase access to critical dental care, reducing caries risk and mitigating its severe health and socioemotional consequences. However, despite the use of evidence-based interventions, approximately 30% of children participating in school-based caries prevention fail to respond to treatment (“nonresponse”), developing new caries and remaining at high risk for subsequent complications. Additionally, the sustainability of school-based prevention is jeopardized due to prohibitive costs, limitations of the professional workforce, and relying on treatments that are not optimized for individual patient needs. We propose to develop and assess personalized, resource-efficient approaches to school caries prevention using adaptive interventions. Evaluated by embedding dynamic treatment regimes (DTR) within a sequential multiple assignment randomized trial (SMART) design, we will explore a number of hypotheses related to adaptive preventive interventions for caries. First, we will test the effects of adaptive interventions on reducing treatment nonresponse using generalized linear mixed models, followed by identifying the best dynamic treatment regime using augmented inverse probability weighted marginal structural models for causal effect estimation. We will subsequently build an optimal adaptive intervention using Q-learning, tailoring treatments to individual patient attributes and responsiveness. We will conclude with a microcosting analysis of the adaptive interventions, ultimately quantifying the intervention costs under different implementation scenarios and taking the first step towards understanding how much payers should reimburse for adaptive interventions. To facilitate completion of these aims, we will implement our adaptive prevention model in the CariedAway network: a multistate community-based participatory research group of schools serving predominantly low-income children that are committed to innovative approaches to caries prevention. This will simplify site and patient recruitment, embeds community engagement into our research process, and has a demonstrated history of both successful clinical care and research. Successful completion of this project will produce resource-efficient adaptive prevention interventions (API) to reduce treatment nonresponse and improve oral health.

NIH Research Projects · FY 2025 · 2025-09

Alzheimer’s disease (AD) is the most common type of dementia without curative medications. Early detection of AD is thus essential for timely intervention and effective treatment development. Multi-view data provides a transformative approach to enhance the understanding, diagnosis, and prediction of AD. This research aims to develop three novel statistical machine learning methods for AD research using multi-view imaging, genomic, and clinical data, focusing on multi-view brain and genomic network analysis, significant differential feature testing, and diagnosis and survival analysis. The specific aims of this proposal include: 1. Develop a novel high-dimensional multi-view data decomposition based on uncorrelated common and distinctive latent factors (C&DLFs) to construct multi-view networks, with application to comparing brain and genomic networks across AD statuses; 2. Develop an optimal false discovery rate (FDR) control method based on a novel semi-parametric hidden Markov random field for high-dimensional spatial multiple testing, with application to identifying significant brain and genomic differences across AD statuses; 3. Develop a highly accurate deep-learning-based diagnosis and survival framework for high-dimensional tabular data, incorporating feature selection and view ablation to enhance cost-effectiveness and data accessibility in clinical practice; 4. Apply the three proposed methods to four large-scale AD-related datasets and disseminate the methods with an open, efficient software package. The three novel methods in our project will undergo rigorous theoretical and numerical analyses. The research team’s extensive expertise in multi-view data analysis, network analysis, imaging genetics, high- dimensional statistics, deep learning, and AD research will make significant contributions to the project’s success. RELEVANCE (See instructions): The research goal is to develop three novel statistical machine learning methods for Alzheimer’s disease (AD) using multi-view imaging, genomic, and clinical data. These methods will focus on multi-view brain and genomic network analysis, significant differential feature testing, and diagnosis and survival analysis. The proposed project will enhance our understanding of neuro-genetic associations in AD and significantly contribute to biomarker discovery, early detection, and improved patient survival in AD.

NIH Research Projects · FY 2025 · 2025-09

PROJECT SUMMARY Respiratory failure is a life-threatening condition affecting millions of individuals in the United States annually. Although mortality rates for respiratory failure have decreased, survivors frequently experience persistent physical impairments and decreased quality of life. Little is known about how factors experienced across the course of critical illness, such as receipt of specific care processes (e.g., early mobilization) or post-discharge setting (e.g., neighborhood environment) influence functional recovery and quality of life. To identify targets for intervention on disparities in recovery from respiratory failure, it is critical to examine the sequential impact of social determinants before hospitalization, care processes during hospitalization, and living environments following hospitalization. The goal of the proposed research is to generate evidence to inform interventions promoting recovery of function following hospitalization with respiratory failure. The central hypothesis of this proposal is that patient-level social determinants, unit-level care processes, and community-level environmental factors produce disparities in long-term outcomes among survivors of respiratory failure. In this research, the applicant aims to: 1) examine the influence of social determinants recorded at hospital admission on functional recovery over time among survivors of respiratory insufficiency and failure; 2) characterize the barriers to providing early mobilization in two high-volume medical intensive care units (MICU); and 3) evaluate the influence of neighborhood environments on long-term recovery among survivors of respiratory failure. The proposed research will take place at New York University (NYU), which provides a robust research infrastructure that supports proposal development and fosters research engagement among NYU faculty and students. Training goals throughout the fellowship award include (1) build advanced methodological skills in state-of- the-art longitudinal modeling techniques applied to complex real-word data; (2) advance capabilities in qualitative data collection and analysis using gold-standard methods from the social sciences and implementation science; (3) develop analytical skills in Bayesian analysis; (4) develop a comprehensive knowledge of key theories, frameworks, and concepts that explain health disparities; (5) strengthen skills in scientific writing and dissemination; and (6) study and practice the responsible conduct of research. Mentorship and supervision from Sponsor, Co-Sponsors, and Contributor will occur regularly throughout the fellowship award to augment all aspects of training. The F31 will also provide crucial protected time to advance toward the goal of a productive career as an independent research career as an epidemiologist of function.

NSF Awards · FY 2025 · 2025-09

With the support of the Chemical Measurement and Imaging Program in the Division of Chemistry, Professor Alexej Jerschow and his group at New York University will advance the development of Zero- and Ultralow-Field (ZULF) Nuclear Magnetic Resonance (NMR) spectroscopy, a cutting-edge chemical analysis technique with applications in real-time reaction and device monitoring. Unlike traditional NMR spectroscopy, which uses large, expensive magnets, ZULF NMR spectroscopy operates in very low magnetic fields, making it portable and cost-effective, while providing detailed chemical insights. This research will improve the sensitivity and precision of this new measurement modality, enabling its use in applications such as monitoring battery performance, detecting environmental changes, and analyzing chemical reactions in real time. This project will foster international collaborations with researchers in Germany and India, training students in advanced scientific fields, including in spectroscopy, quantum mechanics, and computation, and engaging students through partnerships with Pratt Institute to apply spectroscopy in art conservation. These efforts will enhance global scientific networks, support workforce development in STEM, and make chemical analysis more accessible and portable. The project will focus on developing novel sensitivity enhancement techniques, including indirect detection and optimized magnetic field sweeps, to overcome current limitations in ZULF NMR’s resolution and bandwidth. Researchers will investigate relaxation mechanisms at ultralow fields using advanced computational and experimental methods to improve the method’s sensitivity and versatility. These advancements will be validated through applications in battery electrolyte diagnostics, thereby demonstrating ZULF NMR’s potential for in situ chemical analysis. The project’s interdisciplinary approach, combining NMR spectroscopy, quantum mechanics, and sensor technology, will pave the way for portable, high-resolution chemical analysis tools with transformative potential across chemistry and materials science. 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 Awards · FY 2025 · 2025-09

Training the nation’s engineers to understand the economic, environmental, and social context and long-term potential impacts of their work is key to fostering competitive technological innovation. Engineers increasingly face challenges that demand not only technical expertise, but also a deep understanding of how their decisions shape economic, environmental, and social outcomes, an awareness that is essential for advancing responsible innovation, earning public trust, and sustaining national leadership in a rapidly evolving global economy. Yet, despite their importance, these topics remain underrepresented in undergraduate engineering education. When addressed, they are often introduced through upper-level or graduate electives, limiting their reach to a relatively small number of students and restricting their depth of content to mostly introductory levels. Efforts to integrate these themes more broadly into required curricula have proven difficult, and past institutional initiatives have achieved only limited success. One reason for this persistent challenge is that academic change, particularly around curriculum and instruction, is a complex and dynamic process. Faculty, who play a central role in enacting change, operate within institutional systems shaped by competing demands, incentive structures, and cultural norms. Without a clear understanding of what motivates faculty to act, efforts to promote meaningful and lasting curricular innovation are unlikely to succeed. This project will investigate the conditions that reinforce or hinder faculty motivation to incorporate economic, environmental, and social considerations into undergraduate engineering courses, suggesting practical, evidence-informed strategies to support instructional change. These insights will help institutions better prepare engineering graduates to navigate the societal implications of technology and contribute to a resilient, future-ready engineering workforce, thereby advancing NSF’s goals of strengthening the STEM workforce, fostering innovation, and promoting scalable institutional transformation. The project will examine faculty motivation through a systems thinking approach, using New York University (NYU), a large, private research university, as a single-institution case study. NYU offers a rich context in which both top-down (administrative) and bottom-up (faculty-led) efforts to promote curricular change have been pursued. The study will focus on the adoption of the Engineering for One Planet (EOP) framework and pursue two primary goals: (1) to identify factors in the academic system that influence faculty motivation to adopt the EOP framework in their teaching; and (2) to understand how the dynamics among these factors affect faculty motivation to integrate this framework into their curricula. To accomplish these goals, the research team will use qualitative system dynamics modeling. Data will be collected through semi-structured interviews and focus group discussions with engineering faculty, designed to elicit their experiences, motivations, and insights related to adopting the EOP framework in their classes. These data will inform causal loop diagrams (CLDs) that illustrate relationships among institutional factors, such as departmental culture, leadership priorities, tenure and promotion practices, and faculty development resources, and how these factors ultimately reinforce or hinder faculty motivation. Integrating the CLDs will produce a qualitative system dynamics model representing a case-based theory of faculty motivation. This theory will serve as the foundation for actionable recommendations to institutional leaders, educators, and policymakers, identifying potential leverage points that support or inhibit instructional change. The project will contribute to the broader effort to modernize engineering education and ensure its relevance in addressing complex societal challenges. 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 Awards · FY 2025 · 2025-09

Automatically finding and synthesizing information from rich textual sources can support a wide range of use cases across work, education, and personal use. Artificial intelligence systems are already assisting users to fulfill their information needs, from providing encyclopedic facts to answering complex questions that require multiple steps of reasoning. Despite these triumphs, such systems often provide incorrect or outdated information while sounding plausible and authoritative. Furthermore, compared to conversing with domain experts who can answer our questions, interaction with current systems is limited. Instead of engaging in multi-turn interaction with users, asking clarifying questions or follow-up questions, systems mostly take a passive role, aiming to provide accurate information at once. This project envisions interactive systems that critically reason about textual sources to provide high-quality, up-to-date information. This research will advance how language systems interface with rich knowledge sources: parametric knowledge acquired during the language model (LM)’s massive pretraining, documents prepended at inference time, and users who can provide context for their initial input query. The devised systems will model the complexities of real-world scenarios, where users' questions are ambiguous, answers continuously change based on the context of the interaction, and heterogeneous knowledge sources contain imperfect and outdated information. It will develop both data-centric and algorithmic approaches to achieve such goals, (1) expanding the definition of document relevancy to incorporate extra-linguistic contexts, (2) constructing synthetic data to update parametric knowledge and instill multi-document reasoning ability, and (3) developing algorithms to leverage simulated multi-turn interactions. Together, the research will improve how information seeking users interact with systems and how systems interact with knowledge sources. It will enable building systems for wider domains where single-turn interaction over a clean knowledgebase is not feasible. 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 Awards · FY 2025 · 2025-09

This project aims to develop a new artificial intelligence system that works alongside mathematicians to tackle problems that have resisted solutions for nearly a century. Recent advances in large language models can generate creative insights and partial reasoning steps, but they often make mistakes and cannot guarantee correctness. In contrast, traditional tools for verifying mathematical proofs offer rigorous guarantees but are not well-suited for automatically navigating the vast search spaces involved in complex mathematical discovery. This research combines the strengths of both approaches: using AI to explore promising ideas and using formal logic to rigorously verify and refine them. As a high-impact test case, the team will focus on the Hadamard Conjecture, a longstanding open problem with applications in quantum error correction, communication systems, and coding theory. The project will also produce open-source tools, educational materials, and outreach programs to broaden participation in advanced mathematics and AI. The research introduces a unified framework with three key components: (1) a self-evolving reasoning pipeline that uses synthetic data to guide exploration of promising matrix constructions; (2) chain-of-thought and curriculum learning to help AI decompose complex mathematical tasks into simpler subproblems, integrate partial solutions, and generalize from simpler to more difficult problems; and (3) formal verification tools, such as Lean, integrated with preference alignment to ensure correctness and enable a self-improving system guided by symbolic proof signals. Together, these elements form a closed-loop system for scalable, trustworthy proof generation. Anticipated outcomes include new Hadamard matrix constructions, practical software for AI-assisted mathematics, and foundational advances in combining learning and logic for mathematical problem solving. 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 Awards · FY 2025 · 2025-09

How do people derive meaning from sentences? In what situations does the language comprehension process break down? Artificial intelligence (AI) language models such as ChatGPT, which appear to understand and use language as proficiently as humans do, might seem poised to provide potential answers to this question—answers that could not only enrich our scientific understanding, but also help address language processing deficits. But for AI systems to fully realize this potential, they need to process language in a similar way to humans. Many distinct lines of research show that this is not the case. One area where the discrepancy between humans and AI is particularly pronounced concerns temporary semantic ambiguity in language: cases where the first few words of the sentence are consistent with multiple interpretations, and only later in the sentence is it clear which of the interpretations is the correct one. Whereas human readers can encounter significant difficulty when they are required to change their interpretation of a sentence, AI models generally do not. The goal of this project is to better understand the reason for this misalignment between humans and AI models, and explore ways of modifying AI architectures to bring them more in line with how humans process language. In this project, the researchers will benchmark success in their model development by comparing how the models process language to how humans process language using a variety of psycholinguistic measurements. By better aligning human and AI language processing, this research will open up new directions to address long-standing limitations of current AI models, such as their need to train on far more data than human language learners do. In more technical terms, this proposal explores the idea that one key difference between human and machine language processing is that humans: (i) entertain only a small number of semantic interpretations of the input at a time; and, (ii) treat incremental semantic inference as a key goal in language comprehension. This is pursued through three interrelated aims. First, the proposed work will explore the unexpectedly positive correlation between a model’s perplexity and its ability to explain human reading times: put plainly, the better the model is at predicting the next word, the less similar its predictability estimates are to those of humans. Second, it explores whether the human-model misalignment can be alleviated by adopting semantic training objectives and leveraging causal intervention techniques to focus the model’s internal representations of semantic context on a small number of possible interpretations of the input. Finally, human experiments will be conducted to test the predictions of the models on novel psycholinguistic stimuli, with the goal of determining if the proposed modifications successfully bring the models more in line with human language processing. 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.