Massachusetts Institute Of Technology

NSF Awards · FY 2025 · 2025-08

This collaborative project, involving investigators from the Massachusetts Institute of Technology and the University of Washington, focuses on developing new ways to quickly and accurately predict how computer networks perform. Traditional methods for predicting network behavior either simulate each network event in detail, which is extremely slow, or simplify too much, sacrificing accuracy. This project uses machine learning to build fast yet accurate models of network performance. These models will help operators design better, more efficient networks, improving the reliability and speed of services such as online applications, artificial intelligence, and cloud computing. The intellectual merit of the project lies in its innovative use of machine learning to overcome the trade-offs traditionally faced in network modeling. It is structured into three main research activities. First, it develops models that accurately predict critical performance metrics like latency and packet loss using a large corpus of training data. Second, it creates deep learning models capable of forecasting how network performance evolves over time, especially for dynamic applications whose behavior depends on network performance. Lastly, the project investigates how these learned models can be utilized to optimize network configuration in real-time, ensuring networks consistently meet desired performance objectives despite changing conditions. The broader impacts of this project include benefits to both industry and education. Improved network models can lead to significant enhancements in data center efficiency and quality of service, improving user experience, reducing operational costs, and decreasing environmental impact through reduced energy consumption and electronic waste. Industries such as finance, cloud computing, and large-scale AI will particularly benefit from increased efficiency, reliability and real-time adaptability, reducing outages and enhancing resilience against unexpected conditions. For education, the project will generate open-source educational resources that help students and practitioners better understand network dynamics and performance. More information, along with data, code, and results from this project, will be available online at https://github.com/netiken. The repository and related resources will be actively maintained and updated for at least five years beyond the duration of the project. 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 · 2025-08

Air traffic control is a vast, distributed operation that requires the efficient and safe coordination of approximately 45,000 commercial airline flights per day in the United States. Safely managing this volume of traffic requires extensive human coordination combined with sophisticated algorithmic planning. Unfortunately, despite multiple overlapping safety layers, serious safety incidents occur at an alarming frequency. The proposed research combines artificial intelligence with formal methods to improve the safety, reliability, and efficiency of air traffic management. Using a combination of voice recognition, predictive models, and safety checks, the project aims to help detect problems before they occur, such as a misunderstanding causing a future runway incursion or a weather-driven delay that cascades across multiple airports. Beyond air traffic control, the developed framework can enhance other cyber-physical systems that combine human coordination with intelligent automation, improving the safety of firefighting operations, strengthening the resilience of the power grid to adverse events, and advancing national security by preventing human error in military operations. This research aims to advance the theory and application of cyber-physical systems (CPS) by improving the safety and resilience of air traffic management. The project investigates a multi-layered approach that integrates formal verification, robust speech understanding, and data-driven disruption analysis. At the airport level, compositional hybrid automata and signal temporal logic (STL) will be used for predictive monitoring of aircraft trajectories and controller-issued commands, enabling early detection of safety violations. At the interface layer, robust machine learning models will extract semantic intent from noisy, domain-specific voice data to support online reasoning and decision-making. At the regional level, scheduling algorithms will be analyzed under operational uncertainty, using formal sensitivity analysis and probabilistic post-mortem inference to identify failure modes and propose mitigation strategies. While focused on air traffic control, the general methods developed are applicable to a wide range of CPS domains that involve human-in-the-loop operation and algorithmic oversight in distributed safety-critical settings. 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 · 2025-08

NON-TECHNICAL SUMMARY This US-Israel Binational Science Foundation (BSF) collabortive project seeks to establish, test, and explore the limits of a more universal principle to design damage-resistant metals. Metallic materials form the backbone of an enormous amount of US infrastructure. Keeping their properties from degrading is tantamount to ensuring the safety and economic stability of our citizens which this infrastructure supports. Refractory alloys, or alloys which have very high melting temperatures, are particularly promising for seriously demanding applications such as aerospace, defense, medical devices, and nuclear power. However, the nearly infinite possibility of alloy compositions requires a fast and effective way to screen for the most damage-resistant ones. A combination of detailed calculations, high-temperature formation energy measurements, and ultra-rapid, laser-based ultrasound measurements during irradiation are used in parallel. This measures damage as it happens, comparing predictions of which alloys we expect to incur the least damage to reality. The scientific impact of a more universal principle for damage-resistant materials cannot be overstated – such a principle would slow initial material degradation resulting from corrosion, fatigue, radiation damage, or any other source of defects. The broader impact will be far more damage-resistant materials, which can hold their strength and toughness at higher temperatures, under heavier loads, and in more extreme situations. Examples include higher-temperature rocket and turbine parts for aerospace, more robust and longer-lasting medical implants, stronger, lighter armor for critical defense applications, and more radiation-tolerant components for fission and fusion nuclear reactors. TECHNICAL SUMMARY No fundamental, universal principle yet exists to deterministically design materials resistant to atomic-scale damage. Regardless of its source, the defects responsible must originate with a small ensemble of atoms nucleating into a more stable structure, and it is frustration of these stable structures that is the key to preventing defects from accumulating into stable structures which degrade material properties. What has been fundamentally missing from this more universal descriptor of damage resistance is the atomistic process by which point defects and similarly small defect clusters either find their way to similar clusters, forming larger clusters or recombining to lessen the degree of permanent damage. It is hypothesized that potential energy landscape (PEL) roughness, quantified by variance of migration energy barriers (MEBs), stops the rapid, 1D movement of some defects. First, new methods and principles for predicting which alloys in the (WVTa)xTi1-x family, specifically chosen due to availability of molecular dynamics interatomic potentials, will remain in single phase are devised. This is accomplished via thermodynamic calculations and high temperature drop calorimetry experiments. Next, these single-phase alloys are studied with a combination of density functional theory to compute MEB heights and distributions, to compare to in situ ion irradiation transient grating spectroscopy instrumented ion irradiations to confirm whether the core hypothesis is correct. Finally, by linking observed trends in thermo-elastic property changes with thermodynamic calculations of mixing enthalpies and the variance in PEL migration barriers, a simple design principle is articulated and tested to create the most damage-resistant base material composition, upon which other microstructural features (dispersoids, grain boundaries, secondary phases) can be added to further enhance material performance. 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 · 2025-08

The MIT Brain and Cognitive Sciences (BCS) department seeks fundamental knowledge about the brain and mind that can be translated into benefits for society including the diagnosis and treatment of neurological disease. Rigor and transparency are necessary precursors and foundations for these goals. Despite strong shared values and commitment, however, there are daunting obstacles to a sustainable transformation to increased research rigor and transparency. The proposed project is a collaboration between BCS and an external non-profit, the Center for Open Science (COS), that will use a systems-change approach to assess and address obstacles to rigorous and transparent neuroscientific research. In this project, we will translate the effective Theory of Change model developed by the Center for Open Science to the context of a broad neuroscience department. COS’s Theory of Change identifies five necessary and interdependent levels of an effective intervention: [1] infrastructure that makes it possible to do the behaviors, [2] training to make it easy to do the behaviors, [3] making support and behaviors visible to shift community norms, [4] creating incentives to make it desirable to do the behaviors, and [5] re-shaping policies to make the behaviors a required and sustained part of the system. However, this model has not yet been adapted to support change in one of most important subcultures in research -- the academic department. Departments pose a key opportunity for this model, since departments house and are responsible for graduate training programs, and faculty hiring and promotion, and thus the key communities, incentives and policies that guide academic life. The activities of this project are organized to address each of the levels in the Theory of Change. The close collaboration between an academic department and a non-profit culture change organization will produce an effective, scalable solution. By the end of the project period, BCS will have integrated the transparency and rigor-enhancing initiatives into the curriculum and standard practices of the department, and COS will have translated the effective components of the program into an open, scalable suite of products and services that can be exported and adapted to the needs of other departments.

NSF Awards · FY 2025 · 2025-08

Limited availability of raw materials including cobalt and lithium is a roadblock for widespread adoption of Li-ion batteries. One way to circumvent this roadblock is to develop metal-ion batteries beyond lithium. Water-in-salt electrolytes (WiSEs), which are highly concentrated solutions where the salt content is much higher than the water content, are promising electrolytes. These electrolytes offer improved safety, reduced flammability, efficiency, and performance, while also reducing the cost compared to other highly concentrated electrolytes. This collaborative project will use experiments and theoretical modeling to gain a fundamental understanding of the behavior of WiSEs at electrified interfaces. Discoveries in battery electrolytes have the potential to drive economic prosperity through economic growth and technological advancement, while simultaneously improving societal welfare by enhancing safety. The project will support training of two graduate students at the intersection of electrochemical systems and interfacial engineering and will provide opportunities for undergraduates to participate in the research. This project will advance fundamental knowledge of WiSEs at electrified interfaces, focusing on K- and Na-based WiSEs and model electrode materials. The research hypothesis is that tuning the WiSE electrical double layer (EDL) provides a pathway to modulate the heterogeneous electron transfer rate. This project will provide insight into how surface potential, ion specific effects, the amount of solvent and electrode material determine the EDL structure and cluster composition, as well as the screening of electrode charge. Task B will deliver how the species associations and voltage drop at the interface determine the activity and reactivity of each species, and thereby, modulate the interfacial electron transfer. In experiments, the EDL will be studied via attenuated total reflectance–surface enhanced infrared absorption spectroscopy, electrochemical impedance spectroscopy, and potential-dependent force spectroscopy by Atomic Force Microscopy. This will be combined with surface-force measurements using a surface forces apparatus and wide/small x-ray scattering to determine the 3D structure from the interface into the bulk. Mechanistic insight into electron transfer will be gained using in-situ voltammetry with an ultramicroelectrode. For the theory, the current EDL framework will be extended to account for surface effects, divalent cations and redox molecules, calculate species and clusters activities, and incorporate these predictions into an interfacial reactivity model based on the coupled ion-electron transfer theory. The theory will give access to information that is difficult to access experimentally, e.g., molecular configurations at the interfaces, association constants, cluster distribution; this insight will help test and revise hypotheses. The experiments will provide data to validate the 3D interfacial structures, to determine the accuracy of the models, and, in turn, will help improve and tune the theory and 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 · 2025-08

The New England Hardware Security workshop is a regional, one-day conference designed to bring together hardware security researchers, industry participants, and students in order to develop connections and collaborations between these groups. This award will fund student participation in the workshop, reducing the financial burden of attending and thus allowing a wide range of junior students from the region to participate. Student attendees will be able to present their work at poster sessions, allowing them to get feedback from a variety of experts. This award will provide travel support to up to 100 students; the regional nature of the workshop reduces travel costs, accommodating a much wider range of student participants in terms of their career stage and experience in hardware security than most conferences. In particular, the workshop will target early-stage researchers who are not yet attending national conferences, as well as students from disciplinary and personal backgrounds that might not otherwise think to attend a conference on hardware security. This will both increase the chances that participants take up hardware security topics in their future work, grow the hardware research community, and provide participants with ways to connect with potential future employers. 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 · 2025-08

Understanding how the human brain enables us to sense, move, learn, and perceive is a fundamental goal of neuroscience. The brain is made of billions of neurons that communicate with one another through electrical and chemical signals. Neuroscientists rely on fluorescence detecting probes to monitor the activities of neurons using light. However, these probes can be bulky and rigid, causing bleeding and inflammation. Some probe light sources, such as micro-light emitting diodes (LEDs), produce light with a broad spectrum which diminishes the quality of measurements. This project will overcome these issues by developing probes that are much smaller in size, have a softness similar to the brain, and produce light with narrow spectra. The results of this research could enhance our understanding of brain functions and causes for neurological diseases such as Alzheimer’s and Parkinson’s diseases. Furthermore, the research team plans to develop educational materials, including a new course in bioelectronics with hands-on laboratory experience for students. This project proposes to elucidate the effects of beam collimation, compact chip size, and mechanical flexibility of implantable optoelectronic probes on the quality of in vivo deep-brain fluorescence recordings. This will be achieved by vertically integrating inorganic single-crystalline LED and photodiode films obtained via layer transfer technology, distributed Bragg reflectors, and absorption filters to obtain 3D-integrated micro-photometer chips that are less than 1000 times smaller in volume compared to conventional probes, enabling minimally invasive, long-term stable, and high-quality fluorescence recording in deep brain regions. This will improve the signal quality and stability and allow recording in delicate brain regions such as the hindbrain, that preclude the use of traditional bulky/rigid implants. The outcomes of this project will facilitate a fundamental understanding of the neural circuits in these brain regions and advance the knowledge of sensory information processing, complex behavior generation, and neurological disease progression mechanisms in the brain. 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 · 2025-08

To decode and to modulate the emotional aspect of pain Abstract It is well known that pain contains both a sensory-discriminative and an emotional component. While the sensory-discriminative aspects of pain are easily quantifiable, quantitative measures of “emotions” are generally lacking, particularly in preclinical animal models where subjects cannot verbally relay their emotional state. Notably, emotions are all intimately associated with bodily autonomic responses. In preliminary studies in mice, we have measured heart rate responses in acute pain models and discovered that formalin pain dramatically reduces heart rate variability (HRV) while moderately affects heart rate (HR). In aim 1, we will systematically characterize several autonomic responses to identify key autonomic signatures for different mouse pain models. Furthermore, we previously identified general anesthesia activated central amygdala (CeA) neurons (CeAGA) as potent pain-suppressing neurons. Exciting preliminary data revealed that activating CeAGA neurons reduced HR and significantly increased HRV, suggesting that the pain-suppression effect of these neurons is in part due to reversing pain-elicited autonomic responses. In aim 2, we will perform in vivo imaging and functional manipulations of two populations of CeA neurons (CeAGA and CeASst) with concurrent measurements of autonomic responses and behavior in acute and chronic pain models to determine how their activity is causally linked to different patterns of autonomic and behavioral responses. Finally, it is known that past pain experience leads to hypervigilant and anticipatory responses to similar sensory stimuli. Human studies suggest that receiving information about upcoming pain can preemptively elicit autonomic responses. We have established a new pain anticipation paradigm in mice where we can clearly observe the nocebo effect. We also identified pain anticipatory neural signals in the midline thalamic nuclei (MT). In aim 3, we will dissect the role of specific populations of MT neurons in driving anticipatory autonomic and behavioral responses when mice expect upcoming pain and the consequent nocebo responses using in vivo imaging and functional manipulations. We expect that the proposed research will break new ground in our understanding of the emotional aspects of pain. Ultimately, this knowledge can be used to develop non-addictive, non-surgical chronic pain treatments that remove the negative affective component of pain.

NIH Research Projects · FY 2026 · 2025-08

Abstract / Project Summary Microbiome-based therapies have untapped potential to prevent infection and treat disease—augmentation of the microbiome might one day promote colonization resistance against pathogens, treat inflammation, and provide therapeutic metabolites. The grand challenge facing microbiome-based therapies is the unpredictability of engraftment after application; the degree to and conditions under which human microbiomes are permissive to colonization of new strains is poorly understood. This application seeks to define rules of on-human bacterial evolution, ecology, and colonization, towards the long-term goal of determining the species and therapeutic design modalities that have the highest potential for long-lasting probiotic therapy. We use healthy human sebaceous skin as a model system because of its tractability, low complexity, and high similarity across humans, and because states of health are optimal for applying probiotics without instigating an immune response. We propose to profile five abundant species which comprise the vast majority of on and across person diversity in this community, at the high genomic resolution necessary to build rules of on-person engraftment. We hypothesize that our high-resolution, culture- based approach will reveal commonalities across species in terms of their on-person ecology and evolution, with differences that inform the niche of each microbe and species-specific therapeutic strategies. We will leverage a growing collection of longitudinally collected facial swabs from children, adolescents, and adults. In Aim 1, we will build on and across-person phylogenies to understand how strain stability and transmission vary across species and life stages. In Aim 2, we will characterize the extent to which each species’ on-person evolution is dominated by adaptive versus neutral evolution, which will inform the degree to which probiotics will need to be personalized. In Aim 3, we will identify the role to which interbacterial antagonism controls colonization by combining genomic data with high-throughput in vitro screening and modeling. This work will identify common and species-specific rules of colonization in the skin microbiome, and microbiomes in general, laying the foundation for probiotic strategies with high potential for stable colonization.

NSF Awards · FY 2025 · 2025-08

The cells of our body store information about their identity, such as blood, lung or brain, by locking gene expression through the chromatin state. Although common knowledge indicates that the chromatin state locks genes only “on” or “off”, we propose that it can instead lock genes at a wide range of expression levels, thereby enabling analog information storage. The possibility of encoding analog memory in the chromatin state opens a wide range of opportunities, including the possibility of differentiating pluripotent stem cells into sophisticated tissues with gradients of cell types. This could unlock the ability to create organoids that were not possible to create with previous binary memory paradigms as well as new tissues for regenerative medicine. Through the activities of this project, graduate and undergraduate students will be trained in mammalian synthetic biology, chromatin regulation, and mathematical modeling. We will develop new modules for courses taught at MIT and UCSD, use the research material to enrich our mammalian synthetic biology boot camp at MIT, and disseminate our findings broadly to technical and non-technical audiences through local community events. In this project, we propose a model-driven built-to-understand approach to dissect the molecular mechanisms that dictate analog versus binary memory of gene expression. Our project is grounded on the hypothesis that the strength of the positive feedback loop between DNA methylation and histone H3 lysine 9 trimethylation (H3K9me3) determines whether memory is binary or analog. Since the strength of this positive feedback depends on the cellular context, we propose to vary the context by considering different cell lines and promoters and to verify that memory is analog in those instances where the positive feedback is broken. We further propose to engineer analog memory in a cell line where memory is binary by artificially breaking the positive feedback loop through chromatin regulation. This will demonstrate our understanding of the biological rules that make memory analog. We will finally differentiate hiPSCs cell lines engineered with our reporter system to neural stem cells (NSCs) first and then to radial glial cells (RGCs) and monitor gene expression to determine whether and when analog memory emerges. This will validate whether analog memory naturally develops in the neural lineage, where gradients of cell types reminiscent of analog memory have been recently reported. This project is supported by the Systems and Synthetic Biology Cluster of the Division of Molecular and Cellular Biosciences. 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 · 2025-08

Summary ): The application of functional fluorescence imaging to study neural circuit mechanisms in freely behaving animals has become a cornerstone approach in neuroethology. Recently, new fluorescence indicators, such as genetically encoded voltage (GEVI) indicators and new calcium indicators (GCaMP8), have achieved kinetics of milliseconds that can resolve activities as fast as single spikes. However, the current imaging systems used in different model animals have yet to attain the necessary resolution, speed, and signal-to-noise ratio (SNR) to image these indicators at sufficient speed (>1kHz) to resolve spiking events and the temporal coding of neuron ensembles, while accommodating unrestrained behaviors. The key challenges of high-speed fluorescence imaging in freely behaving animals, such as the spatial and temporal resolution trade-offs and the SNR requirements, are fundamentally linked to the camera image sensors. We have pioneered a novel CMOS image sensor with pixel-wise programmable exposures (“PE-CMOS”) to address these fundamental limitations. The PE-CMOS permits independent exposure at each pixel. This feature allows versatile pixel configurations to (1) enhance spatial-temporal resolution at sampling physiological signal while reducing system power and noise levels, (2) boost signal SNR in high-speed imaging with limited pixel exposure, and (3) accurately track cells’ position during brain movement caused by unconstrained behavior, and eliminating motion-induced artifacts. We will develop new image sensors based on the PE-CMOS architecture that will outperform the state-of-the-art image sensors. These are (1) a 600 x 800 pixels sensor with 1 kHz temporal resolution for fluorescence micro-endoscope in freely moving mice. (2) a 600 x 1300 pixel sensor with 11 kHz frame rate and an equivalent 50 kHz max temporal resolution for volumetric imaging of GEVI activity from the whole brain of a freely behaving zebrafish. We will demonstrate these sensors in applications to imaging GCaMP8 and GEVIs at high temporal resolutions while allowing for naturalistic behavior in mice and larval zebrafish — two widely used model systems in neuroscience research. Specifically, we will demonstrate (1) recording and decoding place cell replays at millisecond resolution and capture the spiking activity of fast-spiking interneurons, a task unattainable with current systems limited to frame rates below 60 - 100 Hz, and (2) we will also incorporate the new sensor into our recently demonstrated light-sheet microscopy system to allow whole-brain GEVI imaging of larval zebrafish at > 200 Hz temporal resolution during free tail movement. Achieving these goals will demonstrate the technology’s efficacy at resolving millisecond scale temporal coding structure in cell-type-specific neuron ensembles. It will further establish its applicability in other mammals and small animals commonly used in neuroscience research.

NSF Awards · FY 2025 · 2025-08

Algorithms, cloud computing, and data define the infrastructure of the modern digital economy. The prices users pay for computation, software services, and information must reflect the true value to society of these goods. This is necessary in a market economy both because such prices help foster innovation and because they lead to efficient use of resources. This research analyzes three linked questions: (i) how to price the tokens that govern access to large language models; (ii) how to design cloud-computing contracts that reward sustained but flexible demand; and (iii) how platforms can share and aggregate data while respecting users’ private information. By developing rigorous economic models and translating the results into actionable pricing rules, the project advances the efficient allocation of digital resources, informs regulatory and antitrust debates, and supports workforce development through graduate training and open educational materials. The investigators build and solve mechanism-design models that capture the multidimensional nature of modern digital services. The first component studies a monopolistic provider of a specific kind of AI service (large language models) that sells finite-input, finite-output, and function tokens. This project uses a Cobb–Douglas production framework to derives cost-based nonlinear pricing plans. The team also characterizes welfare outcomes, and identifies empirically testable pricing ratios. The second component models sequential cloud-compute contracts in which users commit to future demand but retain real-time flexibility. This component shows that two-part and budget contracts implement the revenue-maximizing allocation subject to incentive and participation constraints. The third component analyzes data markets in which platforms, advertisers, and sellers trade information about consumer types. The component it compares distribution-platform and advertising-platform regimes, establishes profit and welfare bounds, and derives algorithms for privacy-preserving data sharing. Extensions examine competition among multiple providers, dynamic ticket pricing for compute resources, and heterogeneous buyer populations. Results are derived analytically and, where closed forms are infeasible, with numerical examples based. Outputs include scholarly articles, policy briefs, and open-source code for tariff computation and welfare analysis, enabling researchers and practitioners to apply the findings across sectors that rely on artificial intelligence, scalable computing, and data-driven decision making. 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 · 2025-08

This project develops a new method for comparing human and artificial intelligence (AI) decision-making, generating insight on how to combine human and AI input in high-stakes decisions like medical diagnoses. While researchers can easily observe an AI tool’s prediction process, human decision-making is more complicated. For example, a radiologist’s diagnostic decisions may reflect both their judgments about the presence of cancer and their assessments of the cost of an incorrect diagnosis, but researchers often only observe their final decision. This project develops a method for comparing human and AI decision-making that accounts for complex human preferences and judgements caused by factors that the researcher cannot directly observe. The researchers apply these methods to a radiology setting, using data from a large clinical trial to compare an AI tool’s predictions of lung cancer to radiologists’ diagnoses. This research helps determine the optimal use of AI in high-stakes decisions by answering questions like the following: What valuable information do humans infer that AI assessments miss? What are the intricacies of human preferences during decision-making, and how do these intricacies affect the quality of human decisions compared to AI decisions? This research generates insights that can be used to design better AI-supported decision systems in radiology and other high-stakes settings, leading to public health benefits, a deeper understanding of human and AI decisions, and more responsible use of AI. The project makes three methodological and empirical contributions. First, the researchers show that commonly used approaches for comparing human experts against predictive algorithms are biased in favor of algorithms if the preferences of human experts vary due to factors not observed in the data. Second, the researchers construct bounds on the quality of human expert predictions in the presence of unobserved preference heterogeneity and identify in which specific cases the human expert observes useful private information. Third, the researchers characterize the distribution of preferences across human decision-makers and cases. In an application, the project uses data from the National Lung Screening Trial to compare an AI tool for lung cancer detection against radiologists, empirically quantifying the importance of private information observed by humans that is missed by the AI tool and the extent of preference heterogeneity across cases and radiologists. Using these results, the researchers consider two practical problems: the delegation of cases to humans or AI based on case characteristics, and the design of an automated decision-making tool that aligns AI predictions with the newly determined bounds on human decision-maker preferences. 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 · 2025-08

Rainfall during the Asian summer monsoon (ASM) is a critical water resource for human use, ecosystems, agriculture, energy and industry. The behavior of the ASM varies regionally, and there is a gap in understanding of the variability through time of ASM and the Asian Winter Monsoon (AWM) in mainland Southeast Asia (MSEA), where a greater proportion of rainfall is during the autumn and winter months. This project will reconstruct rainfall intensity on timescales of decades to tens of thousands of years through geochemical measurements of cave deposits from Laos and Vietnam. These measurements, along will climate model simulations, will improve understanding the mechanisms that drive variability of the ASM and AWM, and how this system responds to climate variations through time. This project will use oxygen, carbon, calcium stable isotope ratios and trace elements measured in speleothems, cave monitoring, and hydrogeochemical modeling to reconstruct ASM and autumn/winter monsoon AWM circulation and regional precipitation patterns in MSEA over the last 200 ky. These new data will be synthesized with existing data and isotope-enabled and high-resolution climate model simulations to determine drivers of MSEA hydroclimate variability on millennial to orbital timescales and characterize decadal scale hydroclimate variability and the impacts of coupled ocean-atmosphere dynamics and hydrological extremes on ASM and AWM in different climate states. The project includes support for undergraduate students and PhD students and K-12 outreach events. 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 · 2025-08

Understanding the restless, electrically charged layers that surround Earth's space environment is vital because disturbances in these layers can interfere with navigation satellites, power grids, and radio links. This project will transform Madrigal, a global archive of upper-atmosphere measurements, into an open gateway that anyone can explore. A new virtual assistant powered by artificial intelligence will clarify unfamiliar terms, direct visitors to relevant datasets, and propose short, browser-based exercises that run without requiring special software. Interactive maps will reveal how electron clouds move through the sky, while step-by-step programming notebooks will demonstrate how raw observations are transformed into new information about space weather. Teachers, lifelong learners, and citizen scientists will help design these resources, ensuring the platform remains open, transparent, and continually improving. By lowering technical barriers and welcoming many perspectives, the project will expand the experience of geoscience discovery beyond specialized scientists. This project focuses on enhancing the utility and accessibility of the Madrigal database, a widely used distributed data system for space physics research, through three synergistic initiatives. First, the project introduces an AI-powered virtual assistant built on a retrieval-augmented generation framework that dynamically integrates documentation, metadata, and research literature to guide users in data discovery, analysis, and interpretation. Second, the project establishes a community-driven platform to centralize computational tools and data-analysis notebooks and employs participatory processes to streamline communication between database managers and the space-physics community. Third, the project incorporates capacity-building efforts by developing decision-rich educational modules and innovative visualization resources such as Google Earth based displays. Advanced prompt engineering and large language models integrated into the database will provide real-time, context-aware assistance. Enhanced metadata standards will improve reproducibility and interoperability. Leveraging crowdsourcing and participatory methods, the project will develop a scalable, community-driven framework for creating computational tools and educational resources, with validated evaluations conducted through controlled user studies and pre- and post-assessments. Input from database administrators, researchers, and educators will enable rapid identification of challenges and co-creation of innovative solutions tailored to geoscience infrastructures. 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 · 2025-08

Antibodies in the human immune system recognize and grab specific molecules by changing their molecular structure to present a lock-and-key site tailored for the target. If this could be duplicated in sensors and medical assays, it could lead to new types of medical and environmental tests for human benefit. This project will study how the structures of synthetic molecules, adsorbed on the surface of a sensing nanoparticle to form what is called a corona, can be changed to grab onto specific molecules of importance, such as signaling hormones in humans or toxins in the environment, that indicate stress or diseases. The methods developed in this project will allow scientists and engineers to predict how to synthesize the molecules forming the corona so that they recognize a target for detection. This research has the potential to lead to new types of biosensors and even therapeutics once an understanding of how structure can be tailored to a target molecule. The project will also train the next generation of scientists and engineers in state-of-the-art methods in nanotechnology, and will involve the mentorship of high school students through the HIP-SAT program at MIT to actively encourage increased student participation in STEM. This project investigates the fundamental mechanisms driving interactions between nanoparticle corona phases (CPs) and specific molecular targets. A corona phase is a layer of adsorbed molecules at the surface of a nanoparticle that prevents its aggregation. The project seeks the solution to the longstanding ‘inverse problem’ or how to reverse design the phase so that it recognizes a specific target molecule. Aim 1 focuses on computational methods to predict CP surface coverage and binding constants for specific analytes, leveraging thermodynamic models and Hamiltonian-based approaches to correlate polymer properties with analyte interactions. These new techniques will complement experimental data that are time consuming to generate. Aim 2 seeks to apply molecular dynamics simulations to examine structural and energetic dynamics of analyte binding, providing insights into binding free energy, site availability, and the influence of single-walled carbon nanotube (SWCNT) geometry. This work will produce the first comprehensive modeling approach and workflow for simulating SWCNT CP systems, predicting molecular binding computationally. Lastly, Aim 3 uses these results in a novel droplet-based microfluidic platform for high-throughput validation, enabling precise measurements of cytokine release from T-cells for single-cell phenotyping as a high impact application that builds upon the previous NSF project results. This integrated framework will significantly narrow the design space for molecular recognition sensors, providing practical computational workflows, datasets, and methodologies that will be shared publicly to advance research collaboration and toolkit development for SWCNT CP tailoring. 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 · 2025-07

Project Summary T cells express T cell receptors (TCRs), which recognize antigens presented as peptides displayed by major histocompatibility complexes (pMHC). Recognition of pMHC antigens activates the T cell, enabling immune effector functions leading to the clearance of infected or transformed cells. TCR-T cell therapies are an engineered immunotherapy that take advantage of this biology to engineer T cells to express a tumor-specific TCR to redirect T cell responses against cancer cells. These therapies have achieved objective responses for some cancer patients in clinical trials but have also suffered from limitations in efficacy and toxicity. Since TCRs against many classes of tumor antigens are naturally low affinity, the TCRs used for these therapies often require affinity maturation to achieve efficacy. Unfortunately, this process can also increase the affinity for off-target peptides on healthy cells. Several clinical trial patients have suffered fatal toxicities as a result. Further, affinity maturation is labor intensive, and the resulting mutations are only applicable to one TCR. Cancer antigens are highly diverse, so it is limiting to repeat this process for every new antigen. We propose a new TCR engineering strategy that we hypothesize will address these limitations. We will implement high throughout library screens of TCR constant region variants to select for mutations that improve TCR signaling strength. Because these modifications are not in the TCR antigen recognition domains, they have the potential to improve efficacy without introducing cross-reactivity. Further, they could be applicable to all TCRs, regardless of their targets, which would eliminate the need to re-engineer the TCR for each different antigen. Together, these approaches could identify efficacious, safe, and broadly applicable TCRs as candidates for TCR-T cell therapies.

NIH Research Projects · FY 2026 · 2025-07

Abstract HIV-associated neurocognitive disorders (HAND), substance use disorders (SUD), Alzheimer's Disease (AD), and neuropsychiatric conditions share converging mechanisms of neuroinflammation, immune dysregulation, synaptic dysfunction, and neuronal injury that accelerate cognitive decline. Moreover, SUD and neuropsychiatric disorders co-occur at higher frequencies in persons living with HIV (PLWH) than in the non-HIV population, making it difficult to untangle the influence of these shared genetic mechanisms on progression of the diseases. The Single Cell Opioid Responses in the Context of HIV (SCORCH) consortium aims to address this complex interplay between HIV, substance use disorders (SUDs), and neuropsychiatric conditions. Leveraging the extensive data from SCORCH, PsychENCODE, and Alzheimer’s Disease (AD) datasets, our proposal aims to uncover the genetic and molecular mechanisms driving the interactions between HIV, neuropsychiatric, and neurodegenerative conditions. We employ advanced computational techniques to identify differences in gene expression across disorders, with a focus on identifying shared and distinct drivers of HIV and SUD impact on psychiatric genes and pathways (HAND), and their interplay with Psychiatric disorders (Psych) and dementia/cognitive impairment (DCI). In Aim 1, we leverage single-cell transcriptomics and spatial datasets with advanced machine learning algorithms including contextualized learning, module analysis, and cell-projected phenotypes to discover shared and distinct differentially-expressed genes (DEGs), pathways, modules, and cell types underlying HAND-Psych-DCI. In Aim 2 we also leverage single-cell epigenomics and genetics datasets to infer cell-cell-communication between microglia and neuronal cells, gene-regulatory networks, upstream regulators/TFs, cell-cell communication, and causal drivers, and create a public resource and interactive website for broad data dissemination of all primary, secondary, and tertiary analyses and results for the broader scientific community. In Aim 3 we perform experimental characterization and mechanistic testing in vivo. We develop viral tools for targeting and tracing cholinergic neurons in the mouse basal nucleus of Meynert (BNM), investigate the effects of bidirectional modulation using DREADDs in these neurons, and assess the functional implications of HIV-associated inflammatory signaling. The successful execution of our study will elucidate clinically relevant connections between HIV, neuropsychiatric, and neurodegenerative disorders by examining single-cell transcriptional changes in an integrated experimental and computational framework. This approach enables us to dissect distinct molecular pathways across brain regions and cell types, providing a deeper understanding of the unique and shared mechanisms underlying these conditions. The insights gained will be crucial in guiding the development of new therapeutic strategies.

NSF Awards · FY 2025 · 2025-07

Polymers are encountered in everyday life in products ranging from earbuds to clothing. Their ability to have specific properties (e.g. soft like the silicone in earbuds) arises from the chemical makeup of the molecules (how many carbons, hydrogens, oxygens etc.), their shape (e.g. polymers can have simple linear forms like spaghetti or be circular like a bracelet) and how mixtures of polymers fit together when mixed. There has been a growing trend to synthesize polymers with shapes beyond a simple linear form. Two-dimensional sheet-like polymers are an emerging new class of polymers. Research to date has focused on understanding synthesis of these novel polymers. A better understanding is needed of how to assemble them into complex structures. This project will study how 2-dimensional polymers can be made to assemble into various structures based upon the addition of a second molecular component. A range of added components (both linear polymers and small droplets) will be explored to develop rules for their assembly and to introduce new functionality. A key part of the research entails using specialized cameras and microscopes to take movies of single polymer molecules within these complicated mixtures. The project will train students to enter the workforce for advanced engineering topics combining single molecule manipulations and microfabrication. The project will provide research opportunities for high school teachers and undergraduates. Demonstration kits will be developed which can be used by school teachers to introduce concepts in molecular connectivity in materials. Discoveries from the project will impact the design of future high-performance polymer systems. Single molecule studies of DNA have provided new insights in the fields of soft matter and rheology. There is an emerging interest in polymers with more complex topologies and mixtures thereof which can translate to new material properties. The research team will study clusters formed from 2-dimensional catenated DNAs (called kinetoplasts or kDNA) when mixed with linear DNA or nanoemulsions. The team has recently established kinetoplasts from the Crithidia parasite as a model 2-dimensional polymer system. kDNA are soft, cup-shaped, cell-sized colloidal entities composed of thousands of circular DNA that are topologically interconnected like chainmail. The research program entails: 1) understanding kDNA clustering when mixed with linear DNA, 2) modulating kDNA clusters by using depletants with various properties and 3) studying the nonequilibrium dynamics of kDNA clusters in microfluidic devices. Single molecule fluorescence microscopy combined with bespoke microfluidic devices will be used to study these systems. This award will result in work at the front edge of fundamental studies of soft colloidal assembly/dynamics and rheology of 2D polymer assemblies. New insights will be gained into how 2D polymers can be assembled into superstructures through mixing with linear polymers or smaller colloidal nanoemulsions. The use of DNA of a model polymer will be expanded to two-dimensional colloidal “sheets” called kinetoplasts to understand how catenated DNA demix from linear DNA to form superstructures. The fundamental discoveries from this project will result in contributions to polymer physics texts and journal articles. Results from this work can lead to improved polymer processing, new materials and deeper mechanistic insights into 2D DNA rheology. The research will inform the design of DNA-based superstructures for possible use in biomedicine or “organo-DNA” materials. 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 · 2025-07

The goal of this project is large scale analysis of probabilistic systems known as dimer models. They often have simple descriptions; for example, a randomly chosen tiling of the chessboard with domino tiles is one of them. Yet, as the domain that is being tiled becomes large, such systems exhibit intricate behavior and serve as mathematical models for phase transitions seen in nature. More concretely, the experimentally observed phenomenon of roughening of crystals at temperatures close to absolute zero has similar phenomenology. Mathematical analysis of dimer models is challenging, and it requires a diverse set of sophisticated tools from different branches of mathematics including algebra, representation theory, and algebraic geometry. For a broad range of planar domains, these tools will ultimately make it possible to analyze the large scale behavior at a very fine resolution, offering key insights into a much wider range of probabilistic and physical systems. The proposal is dedicated to dimer models on growing planar graphs with periodically varying edge weights. The overarching goal is to describe limit shapes and bulk/edge/global fluctuations around them for a broad class of domains as the size of the domain grows. Establishing an explicit correspondence between periodic dimer models and Yang-Baxter solvable Solid-On-Solid models of statistical physics is essential, and this bridge will provide access to new algebraic structures that in turn will yield asymptotic results. Another key connection is to the theory of Riemann surfaces, whose geometry is essential to describing the limiting behavior of the models. The zero-temperature limit will also be analyzed, and it will be described in terms of tropical geometry. 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 · 2025-07

Krylov subspace methods are among the most popular classes of algorithms in computational mathematics, particularly when dealing with high-dimensional problems — an increasingly important subject in all areas of engineering, science, and modern technology. Their advantages are simple: they tend to be very fast and relatively accurate. But despite their ubiquity, many questions regarding their behavior and their approximation quality remain unanswered. This project, will further understanding of the behavior and robustness of existing Krylov subspace algorithms and create new algorithms for fundamental computational tasks for which no robust algorithm currently exists. Several of the research projects are suitable for training undergraduate and graduate students. A new computational math seminar at MIT, focused on numerical linear algebra, will draw more students to the field. This project investigates Krylov subspace methods through four specific problems: (1) A complex moment method: Examine the theoretical and practical feasibility of a potential new technique that builds upon theoretical work on the truncated complex moment problem in analysis. (2) Estimating matrix functions: Develop efficient algorithms for computing matrix functions for a variety of graph-related problems by applying recently developed fast Laplacian linear system solvers to rational Krylov subspace methods. (3) Constructing extremal polynomials: Introduce a class of problems that will provide insight into error estimates for Krylov subspace methods for non-Hermitian matrices. (4) Recovering a matrix from its moments: Given a finite moment sequence, when does there exist an orthogonal matrix with those moments? The PI will introduce an algorithm to answer this question for a large class of moment sequences. 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 · 2025-07

This project aims to develop mathematical foundations for understanding and improving graph neural networks (GNNs), which are widely used machine learning models for data with graph structures. Such data arises in recommender systems, molecular modeling and a range of scientific and technological domains. While GNNs have achieved notable empirical success, key theoretical challenges remain, including limited expressivity, suboptimal performance on specific graph types, and performance degradation in deep architectures. This project addresses these challenges by building and analyzing principled models that are both expressive and computationally efficient. The research outcomes will contribute to the development of robust machine learning tools for analyzing complex graph-structured data. Undergraduate and high school students will be actively involved through mentoring and educational programs. The project combines mathematical analysis and model design to advance the theory and practice of graph learning. It will pursue three interconnected directions: (1) developing GNN architectures for solving quadratic programs, a broad class of optimization problems; (2) analyzing the expressivity of subgraph GNNs on graphs with bounded cycles, which frequently occur in applications; and (3) designing new approaches to mitigate the oversmoothing phenomenon in deep GNNs. The work will draw on techniques from graph theory, optimization, and neural network theory. These efforts aim to provide a deeper theoretical understanding and practical advancements in graph learning. 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 · 2025-07

This award funds a research project that combines extremely rich and detailed data with economic theory to study how the burden of market failures is shared in the economy. Market failures prevent resources, such as labor and capital, from being used in the places where they are most valuable. While it is known that market failures are a feature of all economies and contribute to income differences across countries, the question of which types of households are affected most by these market failures is much less researched. This project makes progress on this gap by developing tools to connect and analyze multiple datasets that provide an extensive set of links in the economy between households and firms, between both households and firms and the government, and between the firms themselves. Understanding which groups benefit and which groups lose out from policies that address such market failures is vital to the design of approaches that maximize the welfare of any nation. The research outcomes could be essential for researchers and decisionmakers in shaping optimal U.S. policies and potentially enhancing the wellbeing of households and businesses. This award funds a research project that develops tools to estimate distortions—markups, markdowns, and taxes that prevents resources being allocated to their best uses—at a highly disaggregated level and trace them to individuals through arbitrary trade, employment, and financial networks. This methodology is applied to administrative data, where the research team can map out the flow of goods and money for the entire economy by linking firm-to-firm networks with firm-to-consumer, firm-to-lender, and firm-to-employee networks alongside ownership registries. These methods reveal how the burden of distortions is shared among households belonging to different regions, demographic groups, skill groups, and income levels. In addition, they enable answers to various questions, such as which distortions do the most to compress and expand the distribution of living standards; what trade-offs are observed in approaches designed to improve the impacts of distortions; and to what degree overlapping distortions necessitate our wide-ranging analysis as opposed to focusing on one specific distortion or sector at a time. The theory and methods are developed in a way that they can be applied to any country with similar data. This project advances knowledge in several ways. First, it operationalizes recent theoretical work on general equilibrium models of heterogeneous agents in distorted economic environments. Second, by assembling a complete empirical mapping of economic relationships between agents in an economy, it measures the distributional impact of the main distortions that are present in an economy (those on labor, capital, output, and intermediate inputs) throughout all sectors. This is relevant because studying the impact of reducing distortions in one specific market is influenced by distortions in other markets. Thus, to fully assess the trade-offs of reducing distortions, one needs to go beyond specific distortions and specific sectors. The results have the potential to significantly reshape how economists perceive the implications of market distortions and the policies responding to them. By addressing who bears the costs and who benefits from market distortions, a key theme for decisionmakers, the findings of this research could lead to optimal policies related to market failures and enhance the welfare of the U.S. population. 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 · 2025-07

PROJECT SUMMARY Iron is a micronutrient that is essential for nearly every living organism, and maintaining its homeostasis therefore represents a crucial biological problem across all scales: from cells, to tissues, to the whole organism. We propose that central regulation of local iron supply in the adipose tissue is controlled by VAMs (vasculature- associated macrophages), which are resident tissue macrophages that associate closely with the vasculature and perform several key homeostatic maintenance functions within the tissue. While VAMs and other tissue macrophages have been historically linked to host defense/innate immunity, there is a major gap in our understanding of their diverse functions beyond immunity, and how their dysfunction drives the progression of chronic inflammatory diseases such as type 2 diabetes. Gene expression profiling indicated that VAMs express an array of genes involved in iron handling, including iron import, export, processing, and storage. Our preliminary data shows that VAMs take up significant amounts of transferrin - the main iron carrier in blood serum - and also serve as an iron storage depot by housing significant amounts of intracellular iron. Together, we hypothesize that VAMs are the main regulator of white adipose tissue iron content, constantly monitoring the demand and controlling the local supply of iron to other cells. This study aims to elucidate the tissue-level iron handling functions of VAMs by 1) Identifying the primary iron uptake pathway in VAMs. 2) Ascertaining the role of VAM-mediated iron export and storage in regulation of adipose tissue iron content and systemic metabolism. 3) Uncovering the molecular mechanisms for adaptation of adipose tissue cells to perturbations in local iron availability. To answer these questions, we developed novel mouse models that target tissue-resident macrophages with very high specificity, thus overcoming a major barrier that currently exists in the myeloid field. This study will advance our understanding of the multifaceted roles of our immune system beyond immunity, with a focus on achieving deep mechanistic understanding of macrophage-mediated support of tissue physiology and specifically of tissue-level iron homeostasis. It may inform the biology of several diseases characterized by iron dysregulation, including type 2 diabetes, hematological and neurological disorders, and cancer. The results of this work may influence the development of treatments aimed to restore the proper cellular functions of dysregulated resident macrophages. This is a dissertation project which will be conducted in the lab of Dr. Hernandez Moura Silva at the Ragon Institute of Mass General, MIT, and Harvard, which has all the necessary facilities to perform the experiments described and provides an excellent training environment for the investigation that spans across the fields of immunology, physiology, and biochemistry. The proposed training plan will strengthen the candidate’s preparation to a career in research & development in an industrial sector, where interdisciplinary skills are essential.

NSF Awards · FY 2025 · 2025-07

Large language models (LLMs) have led to significant progress in natural language processing (NLP) and artificial intelligence more broadly, and have the potential to become a broad technology with applications across myriad domains. However, current LLMs rely on computationally expensive architectures and algorithms. This CAREER project aims to develop new methods for efficient, architecture-aware algorithms for language modeling that is expected to make existing LLM applications more efficient, enable new applications, and broaden access. To achieve these goals, this project will develop new methods that span the entire training and deployment pipeline, including: (1) architectural primitives that can overcome the computational inefficiencies transformers, (2) efficient training algorithms that will reduce the amount of resources required to train and finetune LMs, and (3) quantization algorithms along with flexible kernels that can better utilize the computational resources of modern hardware. 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.