During the last military dictatorship in Argentina, also called the Proceso de Reorganización Nacional (National Reorganization Process, 1976-1983), General Jorge Rafael Videla and the other members of the leading juntas sought to subdue the extremist violence that had built up over several decades. Yet, the junta used its mandate to persecute “subversive elements” that it saw as threats to its ideal of a Western and Christian society: political opposition, intellectuals, students, religious minorities, and even apolitical citizens. In this brutal repression operation, also referred to as the Guerra Sucia – the Dirty War – the military government abducted, “disappeared,” tortured, and killed an estimated 30,000 people. After a defeat in the Falklands/Malvinas War, the junta was replaced by Raúl Alfonsín’s elected civilian government in 1983, which implemented a series of policies designed to bring truth, justice, and accountability – core to a concept scholars call “transitional justice” – to Argentina. This effort, which included the creation of a truth commission, the prosecution of perpetrators of repression, and the reparation of victims, was essential for the construction of a new, legitimate, and robust democracy. Through transitional justice, the government sought to repair the social ruptures caused by the Dirty War and ensure that the human rights violations would not happen again. While Alfonsin and his successor Carlos Menem facilitated a period of impunity between the late 1980s and the early 1990s, the pursuit of truth, justice, and accountability continues to this day. While the role of domestic dynamics in this process has been studied extensively, this thesis seeks to uncover the role of other valuable players: transnational non-state actors. These organizations, which transcend boundaries to enact change, are crucial in a world governed by international norms. Thus, my thesis investigates the question: How did transnational non-state actors help promote truth, justice, and accountability throughout the Argentine democratization process? By focusing on three types of transnational non-state actors, regional human rights bodies, international non-governmental organizations (INGOs), and philanthropic foundations, this research highlights the mechanisms used by these organizations to promote transitional justice and democracy from abroad – both during and after the National Reorganization Process. To achieve this, I explore archival material and secondary sources, supplemented by a variety of expert interviews. Ultimately, I argue that by elevating the Argentine human rights situation into the global public sphere, supporting domestic activity, and pressuring both the Argentine and foreign governments to uphold international norms, transnational non-state actors effectively fostered truth, justice, accountability, and democracy.
This research is particularly important in the current era of democratic backsliding, in which elected leaders are consolidating power, eliminating protections for rights, and questioning international norms. I argue that by implementing policies to weaken transnational non-state actors, leaders are tacitly eliminating a fundamental democratic safeguard. By strengthening transnational non-state actors from within, it will be possible to counteract this trend and preserve democracy.

This paper intends to provide a qualitative description of an experimental alternative education program for formerly incarcerated adolescents. In doing so, it intends to construe this school as a site of liberation - in opposition to its former iteration as a correctional institution. It examines the school’s architecture, pedagogical practices, and practices around care to reveal that this institution has been reconfigured to empower students in any number of ways - from meeting their material needs to providing the conditions for immense intellectual enrichment. This ethnographic exploration may be used as a manual for replication or as fodder for thought on what education can do and mean.

The Off-Grid Power Module (OGPM) is a solar panel system designed to power TransAstra’s Sutter Turnkey Observatory (TKO). The module would allow the operation of TKO without being tied to the electrical grid, dramatically expanding the sites that the system can be installed in. The main objectives for the design were for the module to collect and store enough power to operate three TKO telescope modules 95% of the time, for all the components to fit in a standard 20 ft shipping container, and for the system to be able to be feasibly assembled and installed. The resulting design of the system consists of 65 solar panels, 22 lithium batteries, 2 hybrid solar inverters, and a solar array mounting kit. By comparing the theoretical amount of power produced and required to operate the system, and by simulating the probability that the system is able to operate based on weather patterns throughout a calendar year, the selected number of solar panels and batteries in the system were confirmed to meet requirements. The design outlined in this report addresses all design objectives and outlines avenues through which the OGPM could be further developed.

Adolescence is a sensitive period of life shaped by environmental factors, including socioeconomic status (SES) and early adversity. While previous studies have linked early adversity to cognitive and physical outcomes, it remains unclear how these effects vary by income level or whether physical maturation mediates the influence of early adversity on cognition. This thesis investigates how early adversity—specifically prenatal stress and early life stress (ELS)—impacts adolescent cognitive development across income groups and explores whether physical development mediates these relationships. Using longitudinal data from the Adolescent Brain Cognitive Development (ABCD) Study, I compare cognitive and physical development between high- and low-income adolescents and apply linear mixed-effects models and mediation analyses to examine potential pathways. First, I show that high-income adolescents undergo earlier cognitive development, while low-income adolescents experience earlier physical development. Then, I demonstrate that both prenatal stress and ELS are significantly associated with cognitive and physical development in the high-income group, whereas only prenatal stress is significantly associated in the low-income group. This indicates that among low-income adolescents, the influence of ELS may be overshadowed by broader adversities prevalent in low-income environments. Expanding on this finding, I reveal that prenatal stress exerts a stronger influence than ELS on slowing cognitive development and accelerating physical development in the high-income group. Finally, I show that the timing of physical maturation does not mediate the relationship between early adversity and cognitive outcomes, suggesting that the stress acceleration hypothesis cannot be extended to cognitive development. I conclude by proposing that alternative mechanisms—such as emotional development—may better explain how prenatal stress and ELS shape cognitive trajectories, implying the need for interventions aimed at mitigating the long-term effects of early adversity on adolescent cognitive development.

This thesis presents a machine learning-based Wins Above Replacement (WAR) model for NHL skaters, integrating play-by-play and shift data from the 2023–24 and 2024–25 seasons. A Random Forest classifier predicts expected goals (xG) at the shot level, capturing offensive and defensive contributions, while a team-level Random Forest regressor translates performance metrics into win probabilities. Individual player contributions are standardized per 60 minutes, compared to replacement-level baselines, and weighted using feature importances from the win model to compute WAR. The result is a single, context-aware metric that quantifies a skater’s total value in terms of added team wins.

Yellow Fever (YF) caused by yellow fever virus (YFV) has historically been a threat to global health. This risk posed by YFV was highly mitigated with the development of the vaccine strain YFV-17D, but challenges caused by this virus linger despite global vaccination attempts. With outbreaks occurring in urban areas in the past 20 years and the lack of an antiviral for YFV, there is demand for expanding the arsenal of global health tools to fight against Flaviviruses like YFV.

To address this gap, a high-throughput screen was done on the ~75,000 molecules in the Princeton University Small Molecule Library to find a compound that exhibited antiviral activity against YFV while also maintaining low cytotoxicity. The initial screen identified 626 molecules with antiviral activity, of which 43 were determined to be noncytotoxic. While these initial stages of the screen seemed promising, we were unable to identify a molecule that consistently exhibited antiviral activity. Of the 43 molecules that continued to the second round of screening, 5 exhibited a titratable effect where an increase in concentration led to a decrease in viral activity. Those 5 were preliminarily retested but did not validate the original antiviral activity exhibited in the initial screening process.

Due to time constraints, more thorough workup could not be completed to establish firmly whether any of the compounds are suitable for further refinement in structure activity relationship analysis. It is also conceivable that some of these compounds may exhibit antiviral activity against other related flaviviruses. Moreover, the search for an antiviral that acts on YFV can continue by screening other molecule libraries or identifying targets in the YFV replication cycle.

cemented. Yet, the boundaries of what's permissible and impermissible may also fluctuate and develop as a result of these catalysts of change. While there is nothing inherently wrong with progress, unchecked evolution may become domineering and furthermore oppressive. The purpose of this thesis is to evaluate Princeton's progress and, furthermore, the institution's current actions. Essentially, it is to evaluate the fairness, wisdom, and overall accurate judgment of their response against the Palestinian protesters. However, this evaluation will not be done by a student, post-doc researcher, or member of the faculty. Rather, the actions of the past will be called to judge the performance of the present. Using Princeton's past interactions with student-led protests, this paper will examine similarities, consistencies, and, importantly, discrepancies between various eras of protests and the most recent War on Gaza protests. As such, this endeavor will be done through the analysis of four specific eras of Princeotnian protests: The Vietnam era, The South African Apartheid era, The Black Lives Matter era, and finally, the War on Gaza era. In each of the eras, there will be three main connecting variables of comparison. Firstly, the demands of the activists. What was the central issue or problem that the group was protesting against? How do the demands of the group illustrate their values and, furthermore, their relationship with the public and Princeton? The next variable of analysis is the form of expression, i.e., Protests, sit-ins, etc. How did the specific student group express their ideas? Were they confrontational and disruptive? Lastly, the final mode of analysis is the university's response. In other words, how did the university interact with the actions of the student group? Did they meet them with hostility or restraint? Were they deaf or receptive to the demands of the student group? Nevertheless, why are these the three main variables of analysis? These variables are important in understanding Princeton's management of the War on Gaza protests. For instance, if throughout the four cases we find a stark similarity of demands and forms of protests; however, the Institutional responses on the War on Gaza was inherently harsher than other cases , this would provide evidence of inconsistencies of management, hypocrisy, and most importantly discrimination. In other words, let us presume the protestors in the BLM or Apartheid era performed eerily similar protests as their War on Gaza counterparts did and, furthermore, had inherently similar demands. Yet, the Princeton administration treated the War on Gaza protesters in a more brash and harsher way. While this is not on-the-dot evidence for straightforward crimes, it would still pose a problematic situation for Princeton. If an analysis such as this were to happen, a deeper understanding of why Princeton veered away from its methodology of the past would need to be performed. Questions revolving around the notion of what made Palestinian protesters different from Vietnam, Apartheid, and Black Lives Matter protesters if their demands and forms of protesting were similar. Nonetheless, in order to get to that train of thought, the following analysis has to be made. In the next four chapters, each respective era will be analyzed based on this methodology.

Artificial neural networks (ANNs), particularly large language models like GPT-2 and BERT, have demonstrated remarkable capabilities in language comprehension and generation, prompting questions about the extent to which their computational processes resemble those of biological neural networks (BNNs)—the dynamic, adaptive systems of the human brain. This thesis investigates whether ANNs and BNNs share fundamental computational strategies in language tasks, specifically differentiating between task-general computations (predictive coding, context-sensitive representations, hierarchical processing) and task-specific ones (task flexibility and latent cause inference). Using methodologies such as encoding models, representational similarity analysis (RSA), attention-head mapping, and zero-shot generalization, the thesis identifies areas of alignment primarily under conditions of restricted modalities. Notably, predictive coding shows strong surface-level similarity between systems; however, deeper exploration reveals significant divergences in context sensitivity, hierarchical processing, and especially task flexibility—where ANNs fail to replicate human-like interpretation. These divergences underscore crucial limitations in current ANN architectures, highlighting their lack of cognitive scaffolding necessary for genuinely flexible comprehension. This thesis concludes by suggesting future research directions, including integrating multimodal inputs and embedding dynamic memory structures into ANN training, to better assess and narrow the cognitive gap between artificial and biological systems.

Social support is an essential component of emotional and physical well-being and has been implicated in cardiovascular disease risk as well as the experience of daily stress. This study primarily seeks to understand potential differences in receiving versus giving social support and their relationships with cardiovascular disease (CVD), along with an auxiliary focus on the relationship between daily stress and CVD. Through a secondary analysis of the Midlife in the United States study (Wave II), we examined the relationship between daily stress and CVD, the relationship between receiving versus giving social support and CVD, and how these two forms of social support interact in their links to CVD. We found that daily stress does not robustly predict cardiovascular disease, but both receiving and giving social support are associated with a reduced cardiovascular disease risk. However, giving social support ceases to confer protective cardiovascular health benefits on the giver when they receive high levels of social support. Moreover, receiving social support was consistently associated with a lower CVD risk, but the effect was greater when participants were giving a low level of support. These results illustrate that the effects of receiving and giving social support are not necessarily equivalent and should be considered distinctly and simultaneously, rather than in aggregate, in efforts to mitigate cardiovascular disease risk.

This thesis examines the progression of portfolio optimization techniques from traditional (Markowitz and CAPM) to much more computationally advanced techniques such as Machine Learning and LLMs. Using a 15 year dataset of daily S&P500 returns, we show that Long Short-Term Memory (LSTMs) excel at capturing much shorter-term return forecasting compared to Deep Neural Networks (DNNs) which excel at discerning complex, otherwise invisible patterns in the long term and map straight from input data to asset weights. Both approaches surpass classical benchmarks in risk-adjusted performance. Lastly, we introduce a Large Language Model (LLM)–based simulator, demonstrating how ChatGPT can effectively synthesize (e.g., news headline sentiment, policy announcements) into allocation decisions. Our findings highlight the promising future of prompt engineering as well as LLM’s promising ability to combine numerical and textual insight into, potentially, better understood portfolio strategies.

Artisan biochar is a carbon-rich charcoal made from crop waste, commonly produced in soil pits adjacent to the fields where the waste is sourced. Biochar’s potential for carbon sequestration and soil health improvement is widely recognized, so many social enterprises have begun to replace open field burning with biochar production, funded by the sale of carbon credits. However, large-scale implementations face challenges related to cost, labor, and methodological rigor. This paper builds a model then determines an optimal strategy for a manager to inspect a set of n workers who make biochar in a cluster of close-by soil pits, through the duration of a biochar production work shift. Then, the efficacy of this strategy was tested against common strategies for inspection such as randomization and shortest-path decision making, and the results confirm this strategy’s superiority. After that, drawing on real-world data from the “Biochar for Burning” initiative in West Bengal, India, a case study is developed to answer the question: does this optimal strategy make a significant difference when the inspector wants to maximize not work quality, but biochar quality? We revise the model to factor in a dataset of known confounders as noise, run it on the project’s real- world soil-pit coordinate dataset, and determine if the strategy still offers a significant improvement. We found a modest improvement (with respect to alternative models) we expect would be greater if we weighed worker effort level as a more important factor than flame temperature, a known confounder, which in further studies we expect will be verified. Our model consistently produced more reliable results than the alternatives as well.