71446 research outputs found
Sort by
Deciphering Regulatory Mechanisms by which Glucocorticoid Receptor Signaling Shapes Effector Differentiation of CD8+ Tumor-Infiltrating T Cells
No full textCD8+ tumor-infiltrating lymphocytes (TIL) are critical for tumor elimination and undergo a series of states change in the tumor microenvironment (TME), from stem-like T cells (TSL), to cytotoxic effector T cells (Teff), and finally to dysfunctional T cells (Tdys). A prior study indicated that endogenous glucocorticoids (GCs) in the TME shape the effector differentiation of tumor-antigen specific CD8+ T cells, initially restraining the TSL to Teff transition, and subsequently, promoting the development of Tdys. Genetic ablation of the glucocorticoid receptor (GR) specifically in CD8+ T cells improved tumor growth control. However, the mechanisms by which the GR governs the differentiation of CD8+ T cells at different stages remained unclear. In this project, we investigatd how GC signaling impacts on CD8+ T cell state transitions using single-cell RNA sequencing (scRNA-seq) of CD8+ TIL from wildtype (WT) and GR conditional knockout (GRcKO) mice. We found that WT and GRcKO CD8+ T cells differentially accumulate along two distinct differentiation trajectories. WT cells show a delayed transition out of the naïve/stem-like state and preferentially differentiated along a trajectory culminating in dysfunctional T cells. In contrast, GRcKO cells preferentially differentiated along the trajectory culminating in CX3CR1+ effectors and CCL3+CCL4+ cells that share features with tissue-resident memory (TRM) cells and cells present in patients that respond to immune checkpoint blockade (ICB) therapy. Additionally, TCR clonal expansion was more pronounced in the GRcKO cells, particularly in the CX3CR1+ effector cluster. Bioinformatic analysis suggests that this expansion is supported by enhanced glycolysis, indicating a role for GC signaling in regulating the metabolic fitness of CD8+ T cells. We identified Thioredoxin-interacting protein (TXNIP) as a key molecule induced by GR signaling, which can induce cell apoptosis and inhibit glucose transporter 1 (GLUT1) function. Indeed, TXNIP knockout increased CD8+ T cell glucose uptake and enhanced their anti-tumor effects. Our findings provide insight into the effects of GC administration on CD8+ T cell responses and may inform strategies to enhance CD8+ T cell-mediated tumor clearance.Graduate Educatio
Quiet Inclusion: Essays on the Political Economy of Refugee Integration in Africa
No full textThere are more than 120 million forcibly displaced people in the world today. Approximately 85 percent of refugees are hosted in the Global South and more than half are children. In Africa, the average length of displacement exceeds 25 years representing protracted scenarios in which many are displaced for their entire school-age and/or working life. In contrast to countries in the Global North, it is less common for host countries in the Global South to automatically integrate refugees into national services like health and education upon admission. Instead, refugees have historically been segregated into parallel systems managed by non-state providers or otherwise granted limited inclusion. As the scale and length of displacement continues to grow around the world, however, host governments face dilemmas – often characterized by pressure from international actors – about whether and how to integrate refugees into national services, labor markets, and broader social, economic, and political life.
In this dissertation, I present three essays focused on the political economy of refugee integration in national systems in Africa. I frame refugee integration as a distinct political economy challenge shaped by uncertainty, contested responsibility, and divergent incentives across transnational, national, and local interests. Refugees test the limits of state obligation: they are inside national borders but outside political membership.
In the first essay co-authored with Sarah Dryden-Peterson, I argue that traditional political economy frameworks explaining when and why states invest in public services for citizens breakdown when applied to refugees due to uncertainty about refugee futures and the state’s ability to realize returns on investment. By centering “responsibility” as an analytic category, I illuminate how domestic actors, donors, and international agencies negotiate power, incentives, and accountability in ways that diverge from models based on citizen populations. When responsibility for refugee education is contested and refugee futures in the host country are uncertain, I argue that host governments have limited incentives to formalize integration in de jure laws and policies.
In the second essay, I introduce the concept of quiet inclusion to describe a phenomenon in which the de facto inclusion of refugees in public services exceeds what is protected de jure. I measure de facto integration across 24 countries in Africa and find that quiet inclusion can be a strategic solution for host governments facing dilemmas related to integration in constrained policy environments. I use insights from the cross-national analysis to design a vignette survey experiment with elites in Kenya and find that preferences for quiet inclusion hold at micro-level when relevant conditions which characterize the dilemmas elites face are made salient in the vignette scenarios.
In the third essay, I field a framing and conjoint survey experiment with public school teachers and host parents in Kenya and find that the nature and severity of attitudes toward refugee integration varies considerably between Kenya’s two prominent refugee hosting regions. In Dadaab, attitudes toward inclusion are more likely to be driven by prejudice whereas attitudes in Kakuma are more likely to be driven by concerns about the effort and material conditions associated with integration. These findings highlight that integration in policy environments governed by quiet inclusion are likely to be characterized by considerable sub-national variation stemming in part from the attitudes and motivations of local actors who act as street-level bureaucrats to shape the day-to-day boundaries of inclusion.
The frameworks and findings I present have broader relevance for understanding how marginalized populations access public goods in contexts shaped by global inequality. Perhaps most importantly, I find that efforts to formalize integration characterized by quiet inclusion can backfire resulting in less inclusion than what was possible through de facto channels when advocates fail to take seriously the strategic nature of deviations between de jure and de facto policies.Governmen
Neutrinos as a Gateway to the Dark Sector
Full text linkThe existence of dark matter is one of the most important mysteries in modern astrophysics and particle physics. Although strong gravitational evidence supports its existence, the nature of dark matter, including its potential interactions with known particles, is still unknown. At the same time, the origin of the high-energy astrophysical neutrino flux detected by the IceCube Neutrino Observatory remains uncertain. Scotogenic models, in which neutrino mass generation occurs through interactions with the dark sector, are some of the leading theories that motivate dark matter and neutrino interactions. If dark matter and neutrinos interact through a non-zero elastic scattering cross-section, this interaction could leave an observable signal in the high-energy neutrino flux observed by IceCube. The interaction between astrophysical neutrinos and dark matter is strongest in the Galactic Center, where the dark matter column density is highest. This leads to a correlated signal between neutrino energy and arrival direction. In this work, we perform a profile-binned likelihood analysis using ten years of IceCube data. By searching for anisotropies in the energy and direction distribution of the observed neutrino flux, we aim to constrain or potentially detect signatures of dark matter-neutrino interactions. Our analysis offers new insights into the possible coupling between high-energy neutrinos and dark matter, providing a novel approach to understanding both phenomena.Physic
Killing Plasmodium parasites in the mosquito: target identification and feasibility testing for a novel malaria control strategy
Full text linkMalaria is a devastating disease that affects hundreds of millions of people worldwide,
leading to nearly 600,000 deaths in 2023. Plasmodium falciparum parasites are the causative
agent of over 90% of malaria cases and are transmitted to people by the bite of an infectious
Anopheles female mosquito. Vector control strategies that target these mosquitoes have been
instrumental to control efforts, and the extensive rollout of insecticide treated bed nets has
contributed to substantial reductions in malaria prevalence and mortality since the turn of the
century. However, high levels of insecticide resistance have emerged in anophelines, and to-date,
insecticide-resistant mosquitoes have been identified in over 80% of malaria endemic countries.
This widespread resistance jeopardizes the efficacy of current vector control tools and has
contributed to the recent plateau in malaria cases worldwide. New interventions are thus urgently
needed.
In this dissertation, I describe our efforts to develop a novel transmission blocking
strategy that circumvents insecticide resistance by targeting parasites directly with antimalarial
compounds during their development in the mosquito. In Chapter 1, I review the current state of
malaria control and promising new tools, with an emphasis on vector control. I introduce the
concept of mosquito-targeted antiplasmodials and outline key features for their successful
implementation. In Chapter 2, I identify suitable targets for this strategy by performing an in vivo
screen of antiplasmodial compounds in mosquitoes, optimize key inhibitors for improved uptake
and activity, and assess their resistance propensity and transmissibility. In Chapter 3, I test the
feasibility of incorporating hit compounds in bed net-like materials, determine their
antiplasmodial activity in insecticide-resistant mosquitoes, and assess their activity across
sporogonic development. Finally, in Chapter 4 I discuss implications of this work and directions
for further development of this strategy. Altogether, this dissertation significantly expands our
understanding of mosquito-targeted antiplasmodials and demonstrates their potential for malaria
control in the context of widespread insecticide resistance.Biological Sciences in Public Healt
Artificial intelligence and fake reefs: what privative inferences and LLMs tell us about adjective-noun composition
Full text linkThe fact that people understand completely novel phrases is often taken as an argument that linguistic meaning is composed from the meaning of its parts. Thus, a central concern for the study of meaning is how that meaning is composed, especially for open-class content words like adjectives and nouns. This dissertation studies meaning composition and its interaction with context through the lens of adjective-noun modification and the privative inferences that sometimes result (e.g., a fake gun is (usually) not a gun, and a stone lion is not a (living) lion). This dissertation shows that privativity is not limited to a particular class of adjectives, which leads to a new, non-intersective semantics for adjective-noun composition which handles potential contradictions as part of composition. Further, we find that humans and modern large language models (LLMs) can generalize to the inferences of adjective-noun combinations that they have not seen before. Working with LLMs foregrounds the possibility that these inferences could be drawn by other means than meaning composition, such as memorization or analogy. In fact, success on this task is not explained by analogical generalization, as a computational analogy model and a human experiment involving analogy do not yield the expected inferences for all of the dataset. More broadly, the necessary adaptation in experiment design as well as reflection on our standards of evidence feeds into the broader, currently emerging discussion about how to study compositionality in humans and language models alike.Linguistic
Composing intuitive theories across domains and in real time
Full text linkThis dissertation investigates the computational mechanisms that enable people to flexibly and efficiently reason about the world. Building on the framework of intuitive theories as generative programs, I address two fundamental questions about cognitive architecture: (1) how do distinct intuitive theories communicate with each other, and (2) how can these theories be executed efficiently? Using intuitive physics as a primary case study, I present three empirical investigations and two computational models that contribute to our understanding of these questions.
In Chapter 1, I examine how intuitive physics and intuitive psychology interact in moral and causal reasoning. Through computational modeling and behavioral experiments involving visual vignettes of harmful actions, I demonstrate that moral judgments can be modeled as linear combinations of features extracted from both intuitive physics and intuitive psychology. This work provides a computational account of how separate cognitive systems coordinate to solve problems spanning multiple domains.
Chapters 2 and 3 investigate a phenomenon I term ``blended reasoning'', wherein people achieve computational efficiency by flexibly combining simulation-based and abstraction-based reasoning strategies in real time. Using novel computational and empirical methods, including eye-tracking, I show that people strategically arbitrate between different reasoning approaches based on observable features of physical reasoning tasks. This blended approach enables rapid physical reasoning at the cost of accuracy.
Together, these studies advance our understanding of cognitive architecture by revealing how intuitive theories interact and how computational resources are efficiently allocated during reasoning. The findings suggest that human cognition is characterized not only by rich, structured knowledge representations but also by strategic mechanisms for integrating and deploying this knowledge.Psycholog
LegoAI: Auto-Scaling Large Model Training
Full text linkTraining large AI models is computationally intensive. State-of-the-art language and vision models (LLMs and VLMs) often require thousands of GPUs and weeks or even months of training. As models scale to meet the demands of modern applications, efficient distributed training becomes essential, yet remains highly complex. No single distributed training configuration (or training recipe) works across all combinations of model architectures, hardware platforms, and data modalities. Practitioners must explore a vast configuration space through costly trial and error, often building and tuning implementations manually. Even then, out-of-memory errors and sub-optimal performance are common. This complexity is further compounded by the difficulty of synthesizing efficient implementations for selected configurations. Existing frameworks are fragmented across disparate libraries, lack interoperability, and are difficult to maintain, making the development, evaluation, and reuse of training recipes a significant engineering burden.
This thesis introduces LegoAI, a system that transforms distributed AI training into an automated, scalable, and modular process. Given a model, dataset, and hardware configuration, LegoAI automatically selects the optimal distributed training configuration and generates a production-ready implementation that scales to thousands of GPUs. At its core, LegoAI serves as a synthesis engine: it decomposes state-of-the-art training strategies into modular, composable design principles and unifies them within a single coherent framework. In doing so, LegoAI exposes a vast configuration space that comprises not only existing state-of-the-art algorithms but also entirely new designs beyond them. Through high-fidelity simulation, it predicts memory usage and runtime without requiring execution, enabling fast and safe exploration of the configuration space. Finally, for the empirically optimal configuration, it synthesizes an efficient and scalable implementation. In addition to exploring, comparing, and deploying state-of-the-art algorithms, LegoAI enables full-stack research by analyzing and synthesizing entirely new training algorithms derived from the design space through the composition of existing design principles.
We evaluate LegoAI across diverse models, GPU types (A100, H100), and interconnects (InfiniBand, RoCE), demonstrating strong scalability, accurate simulation, and effective policy synthesis. LegoAI achieves speedups of 65.08%, 12.59%, and 30% over optimized baselines on LLaMA 3.1 models at 128, 256, and 512 GPU scales, respectively. It predicts runtime with over 90% accuracy and memory usage with 99.9% accuracy across hardware configurations. To demonstrate LegoAI's research capabilities, we synthesize new memory-efficient training algorithms based on recomputation that reduce overhead by up to 90% compared to baselines, while achieving superior compute–memory trade-offs by matching ILP-optimal solutions and running over 100× faster.
Thus, LegoAI is the first system to unify the synthesis, simulation, and deployment of distributed training strategies, significantly reducing cost, complexity, and uncertainty while enabling broader and more efficient exploration of the large-scale AI training design space.Engineering and Applied Sciences - Computer Scienc
Algorithms and Architectures for Quantum Simulation with Neutral-atom Arrays
Full text linkFast and scalable quantum simulations promise to revolutionize our understanding of complex quantum systems,
In this thesis, we primarily aim to develop algorithms and architectures for leveraging programmable quantum devices, to model complex physical phenomena. As quantum computers can natively capture superposition and entanglement, two key attributes which are challenging for classical computers to accurately model, this approach promises significant benefits in the long run.
We focus primarily on neutral-atom arrays, an emerging experimental platform, although many of our results apply more generally as well. The work is structured into three phases, each progressively advancing the complexity and control of considered experimental hardware, and in parallel the importance and applicability of the considered quantum simulations.
In the first phase (Chapters 1–3), we address the challenge of programming and controlling quantum many-body systems through analog techniques. Analog quantum simulation utilizes continuous control parameters to engineer desired quantum states and dynamics. Chapter 1 introduces novel methods for steering entanglement using quantum many-body scars, harnessing special strongly-interacting dynamics to manipulate quantum entanglement robustly. Chapter 2 advances this approach by showing how Floquet engineering can be used to systematically generate interactions, and how this enables sophisticated control over entanglement and access to novel quantum phases. Chapter 3 integrates these developments into a general framework for programming Hamiltonians into analog quantum simulators with time-reversal capabilities, illustrating the power of programmable analog strategies for simulations of lattice gauge theories.
The second phase (Chapters 4–6) shifts focus to topologically ordered quantum states, known for their exotic long-range entanglement and fundamental significance in condensed matter physics and quantum computation. Chapter 4 presents strategies to enhance the experimental detection and verification of topological order using ideas from the renormalization group. The procedure we develop, order parameters dressed by local quantum error correction, significantly improve the practical observability of these delicate quantum phases. In Chapters 5 and 6, we explore novel techniques for realizing topological phases, by exploiting newly developed experimental capabilities, notably atom reconfiguration, to achieve precise digital control. In particular, we show how to engineer chiral Floquet spin liquids - exotic quantum phases exhibiting robust quantum coherence and non-Abelian excitations — as well as simulations of topological fermionic matter. These advancements not only illuminate foundational physics but also bridge towards robust quantum error correction schemes.
The final phase (Chapters 7 and 8) expands the techniques developed thus far towards the simulation of increasingly complex physical systems relevant to chemistry and materials science. Chapter 7 discusses a general framework for digital quantum simulation of effective spin models, prevalent in condensed matter physics, introducing crucial techniques for engineering and characterizing these Hamiltonians. Chapter 8 extends these insights by proving that fermionic quantum systems, essential for realistic simulations of electronic structures in molecules and materials, can be efficiently encoded into qubits. This advance significantly reduces computational complexity and opens pathways for genuine quantum simulations of chemical systems.
Collectively, these contributions represent substantial progress towards practical quantum simulation with neutral-atom quantum processors, laying critical foundations for future applications in physics, chemistry, and quantum information science.Physic
Opportunity or Desperation: Investigating the COVID-19 Surge in Business Creation
Full text link-Applied Mathematic
Role of SOX4 Transcription Factor in Immune Evasion by Glioblastoma
No full textGlioblastoma (GBM), defined as a stage IV glioma, is an invariably fatal disease with a 5-year survival rate of 5%. One of the defining characteristics of human GBM driving the disease resistance and recurrence is cellular heterogeneity. Specifically, a subpopulation of less differentiated cells displays striking plasticity and resistance to chemo- and radiotherapy. Single cell sequencing identified that two closely related SOXC transcription factors SOX4 and SOX11 are highly enriched in this population.
SOX4 and SOX11, members of the SOXC transcription factor family, play critical roles in embryonic development, stem cell regulation, and oncogenesis. Here, we explore the immuno-evasive function of SOX4 and SOX11 in glioblastoma in both human and murine glioblastoma models. In human gliomasphere models, less differentiated cell state termed neural progenitor cell-like (NPC-like) cell state—where SOX4 and SOX11 are most highly expressed—exhibited immuno-evasive phenotype, namely resistance to interferon stimulation, and low cell surface expression of MHC class I and interferon receptors. Genetic knockdown or pharmacologic inhibition of SOX4 in these models resulted in a dramatic reduction of the NPC-like cell state and shift towards mesenchymal-like (MES-like) cell state. SOX4 knockdown gliomaspheres showed increased gene expression of immune-related genes and increased sensitivity to interferons, suggesting intrinsic immune activation. Type I and II interferon treatments also directly suppressed SOX4 protein levels and reduced the proportion of NPC-like cell state, linking inflammatory signaling with cell state transitions.
Genetic disruption of Sox4 in mouse CT-2A and SB28-Ohlfest glioma cell lines revealed enhanced expression of interferon-stimulated genes and increased sensitivity to type I and II interferons, consistent with intrinsic immune activation. In the SB28-Ohlefst model, these effects were amplified in Sox4/Sox11 double knockout, with evidence of increased vulnerability to T cell-mediated killing in vitro and greater selection against dOVA model antigen-expressing cells in vivo. However, the lack of consistent in vivo response between experiments precluded us from demonstrating overall survival advantage of animals bearing Sox4/Sox11 double knockout tumors.
Collectively, these findings identify SOX4 as a key regulator of immune evasion and cellular identity in human and murine glioblastoma. SOX11 is also relevant but has a less central role. Targeting of SOX4 could be used to decrease heterogeneity of GBM cell states and increase the sensitivity of GBM to immune attack.Immunolog