71446 research outputs found
Sort by
Regularization in Reinforcement Learning: Equivalences and Novel Methods
Get PDFReinforcement learning (RL) is a powerful framework for sequential decision-making, with applications ranging from robotics to healthcare. However, in real-world settings, such as mobile health (mHealth), RL faces challenges due to limited data and the need for generalization beyond observed experiences. Regularization -- a set of techniques that constrain model complexity to prevent overfitting and promote generalization-- plays a crucial role in overcoming these challenges. This dissertation critically examines existing RL regularization methods, uncovers novel connections between them, and introduces new approaches inspired by the challenges of mobile health studies.
One focus of this work is establishing theoretical connections between existing regularization methods. We prove that discount regularization produces the same optimal policy as a Bayesian prior on the transition function and a penalized Q-function, and is also equivalent to a truncated lambda return. These relationships reveal underlying assumptions and limitations of discount regularization.
This work also focuses on introducing novel regularization methods. First we introduce a state-action-specific regularization method that mitigates the limitations of discount regularization uncovered in our analysis. We also propose a novel Bayesian hypothesis testing-based regularization approach that leverages prior study data to improve learning while adapting to differences between the environments of the prior and current studies. This is particularly useful in mobile health applications where feedback is sparse and exploration is limited.
Through theoretical analysis and empirical validation, this dissertation advances the understanding of RL regularization methods and introduces new techniques that enhance generalization in data-constrained environments. These contributions provide a principled foundation for improving RL applications in healthcare and beyond.Engineering and Applied Sciences - Applied Mat
The Effects of News Media Bias on Election Outcomes: A Comprehensive Study of Interpretive-Framing, Priming and Agenda-Setting
Get PDFNews media has the ability to create biases and influence audience perceptions by emphasizing certain elements of a narrative and minimizing others. However, different news biases have different degrees of effectiveness in encourage (or discourage) broader political participation and affecting voters’ political preferences in different geographic regions and electoral periods. This research studies the appearance and the timing of various media bias (using interpretative-framing, priming and agenda-setting techniques) before and after presidential elections in local political news reporting in all 50 states from 1980 to 2021. The empirical results show that the use of liberal-leaning priming technique used in the news increases the probability of electing Democratic presidential candidates while reducing the chance of Republican candidates. On the other hand, the use of conservative-leaning priming and agenda-setting techniques reduces the probability of electing Democratic candidates. In terms of voter participation, left-leaning interpretive-framing reduces the voting rate of Republican voters and left-leaning priming increases the rate of Democrat voters. Finally, the appearance of conservative-leaning agenda-setting technique is associated with a lower voter participation in the Democratic Party.Extension Studie
Unveiling a novel role for macrophages in maintaining and restoring lung epithelial homeostasis
No full textDue to its constant exposure to the external environment, the respiratory tract serves as a
primary entryway for pathogens, including respiratory viruses. Infection causes lung injury
through both direct viral infection of epithelial cells and collateral damage from antiviral
immune responses. To preserve barrier integrity and maintain gas exchange, tissue repair
mechanisms must function even during active antiviral immune responses. Macrophages, innate
immune cells present in all tissues, are essential for homeostasis, host defense, and tissue repair.
Given their strategic positioning and functional versatility, we hypothesized that lung
macrophages, in the context of infection, provide signals to limit the detrimental effects of
antiviral responses and promote epithelial repair in the damaged lung. Our work identified
Oncostatin M (OSM) as a key cytokine with distinct roles in maintaining lung homeostasis and
responding to viral challenges. Single-cell transcriptional profiling and genetic mouse models
revealed lung macrophages as a major source of OSM. At baseline, Osm-deficient mice
exhibited altered alveolar type II (ATII) epithelial cell states, a defect reversible with daily
administration of OSM to the lung for one week. Following challenge with viral stimuli, Osm-deficient
mice displayed elevated type I interferon (IFN-I) levels, more severe lung damage, and
increased mortality, indicating that OSM is required for host survival. Notably, Osm deficiency
resulted in a significant loss of ATII cells following challenge, a hallmark of alveolar injury. While
IFN-I is crucial for antiviral defense, its dysregulation leads to exacerbated inflammation,
cell death, tissue damage, and impaired repair. Upon challenge with viral stimuli, blockade of
IFN-I signaling protected macrophage-specific Osm-deficient mice from morbidity and
mortality. Moreover, OSM delivery to the lungs of these mice was sufficient to reduce morbidity
and restore ATII cell numbers, even in the presence of elevated IFN-I levels. Consistent with
these findings, in alveolar organoids, OSM treatment promoted organoid formation,
counteracting IFN-I’s inhibitory effects on cell growth. Thus, OSM functions as a key growth
factor that accelerates repair and preserves epithelial integrity alongside antiviral defenses.
Altogether, these findings identify OSM as an essential macrophage-derived factor that
maintains homeostasis of lung epithelial cells and promotes their proliferation to overcome IFN-I-mediated
immunopathology.Immunolog
War in Wôbanak: Environmental Histories of the French and Indian Wars, 1675-1763
No full textIn “War in Wôbanak: Environmental Histories of the French and Indian Wars, 1675-1763,” I argue that a century of conflict fought in northeastern North America can be explained by understanding different perceptions and relationships brought to bear on the natural world by members of the Wabanaki Confederacy, officials and soldiers of the British Empire, and English (descended) settler colonists. In Wôbanak, the Dawnland, the first place the sun rises each day in North America, stretching across what most maps now call Maine, Vermont, New Hampshire, Quebec, and the Canadian Maritimes, the people of the Wabanaki Confederacy, the Abenaki, Penobscot, Passamaquoddy, Maliseet, and Mi’kmaq made their home. And for the better part of one hundred years, they defended that homeland in the face of colonial and imperial expansion. While colonists and imperial officials insisted that the natural world could be commodified dominated, and extracted, Wabanaki people saw a space teeming with life and with relationships. By viewing the roles trees, non-human animals, agriculture and placemaking, and even pathogens played in these conflicts—and how differing ecologies shaped and were in turn shaped by them—these conflicts appear as environmental events. With the ascendance of the British Empire and the end of this story, their victory is a pyrrhic one as they are subsumed by settler colonists whose own environmental logic set the stage for contemporary environmental disaster.Histor
Assessing Instructional Explanations for Mathematical Procedures at Scale Using Animated Teaching Simulations
No full textStudent mastery of elementary mathematical procedures is foundational to learning in the discipline and to success in more advanced mathematics. Prior studies suggest that classroom instruction often focuses on the steps of procedures without also providing support for their meaning, for instance by emphasizing place value or justifying steps. However, studies on this topic are either outdated or limited in scale. In response, we analyze 324 teachers’ spoken instructional explanations in reaction to 6 animated teaching simulations covering 3 teaching tasks—explaining a procedure, addressing student confusion, and summarizing a nonstandard student method. An analysis of these data reveals that teachers primarily focus on the steps of procedures except when summarizing nonstandard student methods. Results provide clues about the nature of US classroom instruction and offer a new tool for evaluating the impact of efforts to change that instruction.Accepted Manuscrip
The Five Gs for Teaching Statistics: Greek, Graphs, Grammar, Gadgets, and Games
No full textTeachers of quantitative research methods face multiple challenges in advancing quantitative proficiency, including varying levels of comprehension, engagement, and belonging among incoming students. Consistent with consensus in the learning sciences, I introduce a mnemonic that can remind teachers and students to engage around multiple representations of statistical concepts. “The Five Gs” stands for Greek, Graphs, Grammar, Gadgets, and Games. The first three Gs, Greek, Graphs, and Grammar, refer to mathematical, graphical, and written representations, respectively. The last two Gs refer to Gadgets and Games, which I use to refer to tangible objects and friendly competitions, respectively. I locate the Five Gs in relevant learning sciences literature and demonstrate their application through teaching correlation coefficients. I conclude with guidance for judicious implementation in quantitative methods instruction.Accepted Manuscrip
Surveillance Capitalism in Fragile Democracies: Defending Journalism from Dual Domination
Get PDFThis paper investigates how surveillance capitalism enables a convergence of state authoritarianism and digital monopolies, posing an existential threat to fragile democracies by examining its corrosive effects on the news media in Turkey and Hungary. Drawing on document analysis and interviews with 24 journalists and media executives, it argues that a system of “dual domination” has emerged, wherein the state-led construction of an “economy of domination” through media capture and legal repression is amplified by the market-driven “instrumentarian power” of Big Tech platforms like Google. This symbiotic relationship creates a uniquely corrosive environment that systematically dismantles the free press, pollutes the information ecosystem with propaganda and disinformation, and makes independent journalism economically unviable. The analysis, framed by Shoshana Zuboff’s theories, concludes that this dynamic constitutes a “Second Tragedy of the Commons” and a slow-motion coup des gens, seizing sovereignty from the public and eroding the “right to the future tense.” In response, the paper proposes a framework for solutions grounded in Mathias Risse’s concept of justice, advocating for two pillars of action: first, the establishment of journalism as a legally protected global public good, and second, the formation of a global civic alliance to enforce structural and regulatory reforms that hold tech platforms accountable for their role in demo-
cratic backsliding.Author's Origina
JONES-19: A Cultural Image Dataset Based on The Grammar of Ornament
No full textWe introduce JONES-19, a high-quality image dataset documenting 1,901 ornament designs belonging to nineteen human cultures. The images and their annotations are based on an open access archive of The Grammar of Ornament (London, 1856), by Owen Jones. The dataset poses numerous challenges as a benchmark for computer vision classification tasks and for research at the intersection of machine learning and art and design: a small sample size, image samples of human-designed artifacts rather than common objects in-context or natural scenes, imbalanced class distribution, and image distinctions based on fine details involving line patterns, reliefs, and colors. As a design-inspired dataset, JONES-19 can serve as a benchmark for various research fields, such as visual recognition, data-efficient learning, art-historical research, architectural style analysis, and cultural heritage. This paper describes the curation of the JONES-19 dataset and reports a baseline classification benchmark that evaluates the suitability of the dataset for training classifiers and exposes insights into inter-class relationships–particularly cultural similarities–by examining patterns in misclassification errors. The dataset and its accompanying documentation are available at: https://huggingface.co/datasets/harvardseas-cultural-ornaments/JONES-19.Harvard Data Science InitiativeComputer ScienceHistory of Art and ArchitectureAccepted Manuscrip
Harvard Film Archive Program Calendar January – May 2025
No full text2 MOON MOVIES: APOLLO 11 AT 50
june 8 � august 3
4 EXTREME CINEMA. THE ACTION DOCUMENTARIES OF KAZUO HARA
june 10 � june 23
CALENDAR
6 JUNE
7 JULY
8 AUGUST
9 THE COMPLETE HOWARD HAWKS
june 14 � august 18
21 CINEMA OF RESISTANCE
june 17 � august 26
22 JOAN TEWKESBURY�S OLD BOYFRIENDS august 23 � august 25
23 DARK WATERS. ALL-NIGHT MOVIE MARATHON
august 31 � september
