50107 research outputs found
Sort by
Fair Algorithms for Sequential Learning Problems
No full textMachine learning (ML) systems are increasingly used to automate decision-making across a range of domains. These systems can deliver substantial benefits, but they can also propagate or exacerbate biases that may arise from training data, system design, optimization objectives, or the contexts in which they are deployed. If left unaddressed, these biases can lead to harmful outcomes. Fairness becomes especially critical when such outcomes disproportionately affect certain communities. While a growing body of research aims to mitigate unfairness in ML systems, many challenges remain.
This proposal addresses three problems in fair ML. The first focuses on promoting fairness in the offline contextual bandit setting, where an algorithm must use historical data to, with high probability, find a fair policy. Importantly, different application domains may require different fairness criteria, and an additional challenge is to allow for the specification of fairness from a broad class of definitions. To address these challenges, we introduce RobinHood, a fair bandit algorithm. Our theoretical results confirm that the probability RobinHood returns an unfair policy is less than a user-defined constant. Moreover, we show empirically that RobinHood can satisfy various definitions of fairness across three domains, including a user study with an automated tutoring system.
The second problem involves mitigating long-term unfairness in the classification setting. Most definitions of fairness in classification are myopic and do not directly account for how model predictions impact the well-being of different communities over time. Towards this, we introduce ELF (Enforcing Long-term Fairness), an algorithm whose theoretical and empirical results demonstrate its ability to successfully provide models that satisfy long-term fairness goals.
The final challenge concerns promoting fairness in online resource allocation, an application domain that is often formulated as a sequential learning problem. In this setting, a scarce resource must be distributed across groups over a sequence of rounds, with the goal of optimizing group welfare subject to fairness criteria. We consider a definition of fairness that generalizes principles of equality of opportunity, which asks that the probability an individual receives a resource is approximately independent of their group affiliation. We develop algorithms that identify welfare-maximizing allocations under this fairness definition in both non-noisy and noisy settings. In the non-noisy case, we provide theoretical guarantees, including bounded fairness regret.This research was supported in part by gifts from Adobe, Meta Research, and Google, and by the National Science Foundation under grants CCF-1453474, IIS-1753968, CCF-1763423, and 2018372, including an NSF CAREER award to Emma Brunskill. It was also sponsored in part by the U.S. Army Research Laboratory under Cooperative Agreement W911NF-17-2-0196 (ARL IoBT CRA).Doctor of Philosophy (Ph.D.
Transición precaria: resistencia cultural e intervención queer en la Transición española (1973-1986)
No full textThis dissertation examines queer cultural production during the Spanish Transition (1973-1986) as forms of resistance that challenged hegemonic narratives of peaceful democratization. While official accounts have presented this period as an exemplary process of moderated political change characterized by institutional agreements and reconciliation, this research demonstrates how queer subjects created alternative cultural spaces that defied both Francoist norms and emerging democratic frameworks. Through analysis of three key figures—intellectual Alberto Cardín, cartoonist Nazario Luque, and amateur film collective Els 5 QK's—this study shows how marginalized creators developed strategies of cultural intervention that exceeded traditional underground/mainstream categories.
Grounded in queer theory and Iberian Cultural Studies, this research positions queer cultural practices not as marginal phenomena but as constitutive elements of broader social transformation during the Transition period. The dissertation contributes to the field by offering a critical rereading of the Spanish Transition, demonstrating how cultural resistance from the margins participated in reconfiguring Spanish social imaginaries. By recovering silenced voices and examining diverse practices including essays, comics, amateur cinema, and performance, this research reveals previously invisible genealogies of queer resistance that complicate our understanding of democratization processes in contemporary Spain.Doctor of Philosophy (Ph.D.
Passive cavitation mitigation on hydrofoils via porous media: A comparative study of LES and RANS models
No full textThis study numerically investigates the use of porous media as a passive strategy for mitigating cavitation on a NACA 66 (MOD) hydrofoil subjected to unsteady two-phase flow. Employing the Volume of Fluid (VOF) method alongside the Schnerr–Sauer cavitation model, simulations are performed at two cavitation numbers (σ = 1 and 0.7) for both 2D and 3D geometries. The porous region, characterized by a porosity of 0.95. and spanning one-third of the suction side, is represented as a momentum sink using Ergun's equation in ANSYS Fluent. Three turbulence models are utilized in the 2D simulations: Large Eddy Simulation (LES), realizable k–ε, and k–ω SST turbulence models to evaluate their predictive performance. For the 3D simulations, the LES turbulence model is employed to capture three-dimensional flow and validate the effectiveness of the porous media. The treatment of the porous surface influences cavitation dynamics by stabilizing the boundary layer, minimizing sharp pressure gradients, and limiting the growth of cavitation bubbles. Contour visualizations of the pressure coefficient, velocity magnitude, and vapor volume fraction indicate that the porous layer reduces negative pressure near the wall, lowering local suction intensity and delaying cavitation inception. As a result, the formation of re-entrant jets and aggressive cavity shedding, which are dominant in non-porous setups, is mitigated. Additionally, distributions of turbulent kinetic energy (TKE), z-vorticity, and baroclinic torque illustrate that the porous media diminishes vortex strength and wake instabilities. Spectral analyses of drag and lift coefficients through Fast Fourier Transform (FFT) reveal a notable reduction in high-frequency components for the porous case, demonstrating its ability to mitigate unsteady hydrodynamic loads. Among the tested configurations, a trailing-edge porous layer with a porosity of 0.95 provides the most balanced performance, offering the best compromise between hydrodynamic efficiency, cavitation suppression, and flow stability for practical operation. Collectively, these findings highlight the potential of porous integration as a robust and energy-efficient approach to enhance hydrofoil stability and cavitation resistance in marine systems
GARLIC MUSTARD AND GLOBAL CHANGE: MEASURING TRAIT RESPONSES IN ALLIARIA PETIOLATA TO INTERACTIVE EFFECTS OF SOIL WARMING AND INCREASED NITROGEN
No full textThe concurrent ecological challenges posed by a rapidly changing climate and the negative impacts of invasive species require further insight into the mechanisms by which specific global change factors affect invasive plant performance and potential spread. Current evidence supports the persistence of plant invasions specifically via climate change, due to shifting background abiotic factors that may facilitate the further expansion of species invaded ranges. However, context dependent responses to specific global change factors across populations means that quantifying geographic and population-level variability within invasive plant responses to future conditions will help us to better understand the mechanisms underlying predictions of future spread. The objective of this dissertation is to enhance our current knowledge of invasive plant responses to global change and demonstrate the value of this information supporting land management of invaded sites. This dissertation provides insight into the potential responses of the widespread invasive plant Alliaria petiolata (garlic mustard) to the concurrent global change factors soil warming (SW) and increased levels of soil nitrogen (N), that are expected to shift in the region. Specimens used in the experiment were collected from distinct populations of garlic mustard (Massachusetts, USA) and grown under SW conditions of +5 ℃ above ambient temperatures, N addition of 5 mL of ammonium nitrate (NH4NO3), and a combined treatment of both SW+N together. The first chapter is a narrative review evaluating the current knowledge of garlic mustard’s responses to global change in relation to our understanding of invasive plant responses to global change more broadly. The second and third chapters showcase the results of original experiments measuring garlic mustard trait responses to warming and increased nitrogen conditions. In chapter 2, I measured the growth of garlic mustard seedlings from three different populations in Massachusetts and observed effects from the SW+N and N treatments, population, and treatment x population interactions on all traits measured except root:shoot ratio. In chapter 3, I measured traits associated with growth and reproduction in adult garlic mustard from a population in Concord, MA. The results demonstrate that SW+N increased plant height. While the N treatment alone also increased the number of leaves, branches, and siliques. Sampling period and an interaction between treatment x sampling period was also significant for certain traits. Our results suggest increasing regional SW+N as global change progresses may support the continued growth and development of garlic mustard seedlings. However, N is also a crucial factor at both life stages. As observed in other studies, climate and global change factors may not enhance reproductive output in garlic mustard. The role of regional environmental conditions may be one of the primary drivers for this trend, but it needs to be evaluated further along with other environmental parameters across life stages, populations across broader spatiotemporal scales, and population-level demographic processes.New England Botanical Society Graduate Research Award
OPD Pre-Dissertation Grant
UMass Return to Research Grant
OEB Research GrantDoctor of Philosophy (Ph.D.)2026-03-0
Accelerating Fokker-Planck Simulations by Substituting the Moment Closure with a GPU-Native Deep Neural Network
No full textParticle-based Fokker-Planck (FP) models represent a high-fidelity method for simulating rarefied gas dynamics, but they suffer from a severe computational bottleneck: the “closure problem.” This step requires the expensive, cell-wise calculation of high-order moments and the solution of a 9 × 9 linear system at every simulation time step. This
paper introduces a new computational methodology designed to eliminate this bottleneck by substituting the physics-based solver with a Deep Neural Network (DNN) surrogate
deployed via a novel, high-performance strategy. Our workflow makes a critical distinction between a complex offline training phase (where a 16-256-256-256-256-9 DNN is trained) and a lightweight online inference phase. Crucially, for online deployment, we avoid all framework overhead and I/O bottlenecks by extracting the raw parameters (weights and biases) and executing the model’s forward pass as a simple, batched matrix-multiplication function written natively in CuPy, ensuring all operations remain on the GPU. We validate this approach through a rigorous, multi-stage test campaign. First, for 1D Couette flow, a model trained on a Knudsen number sweep (Kn ≈ 0.0015−0.3) demonstrates outstanding accuracy in both interpolations (Kn = 0.05 and Kn = 0.09) and significant extrapolation (Kn = 0.7). To test fundamental generalization, we deployed this 1D-trained model to the 2D cavity geometry. This test yielded excellent agreement for velocity and density structures but produced minor, localized errors in the temperature field, confirming that representative multi-dimensional data is required for full thermal accuracy. Consequently, a robust 2D cavity model, trained on a lid velocity sweep (50 m/s to 600 m/s), proves capable of extreme extrapolation, accurately predicting the complex, high-energy physics of a hyper-velocity 800 m/s case. The primary finding of this work is a fundamental shift in the computational paradigm for this method. Performance benchmarks show a 1.63x–1.73x speedup, but critically, a strong-scaling analysis proves this acceleration reaches the theoretical maximum predicted by Amdahl’s Law. This result provides a definitive insight: the GPU-native surrogate is, for all practical purposes, “infinitely fast” (zero-cost) relative to the remaining tasks, and the true computational bottleneck has been decisively shifted from the physics solver to the particle moment-gathering process itself
Optimized Development of Housing Neighborhoods: Energy Efficiency and Thermal Comfort Potential of Residential Block Morphology in Harsh Climate Regions of United States
No full textThe escalating impacts of climate change have intensified extreme weather patterns, producing scorching summers and severe winters across many regions of the United States. These shifts not only drive higher building energy consumption but also degrade outdoor thermal comfort, undermining efforts to foster socially engaging neighborhood environments. This research responds to the pressing need to reconsider residential development by optimizing building morphology to balance energy efficiency and outdoor comfort. Although the vision of socially vibrant neighborhoods has long been pursued, harsh climates diminish outdoor livability while simultaneously straining indoor comfort and energy demand.
To address this challenge, the study emphasizes the strategic application of building block forms, integrating advanced simulation and analytical tools—ENVI-met, DesignBuilder, BEopt™, and RayMan—to optimize both individual buildings and neighborhood-scale morphology. The thesis investigates the thermal and energy performance of various low-rise residential development patterns in both hot and cold U.S. climate regions, aiming to support indoor thermal comfort while also enhancing outdoor environmental quality for residents.
The inquiry is structured around two central research questions: (1) To what extent can building development morphology improve thermal comfort and energy efficiency in low-rise residential buildings? (2) What specific morphological features contribute most to these improvements in harsh climates? To answer these, the study adopts a multi-method approach, including a review of literature on block morphology and thermal comfort standards, parametric analyses of building and block forms using DesignBuilder and BEopt™, field measurements for model validation, and case studies of neighborhood microclimates in hot and cold climates assessed with ENVI-met and RayMan.
The ultimate goal is to propose a holistic framework for neighborhood design in diverse climates, advancing energy efficiency and thermal resilience at both the building and block levels. The findings offer practical insights into how morphology can be harnessed to enhance thermal comfort while reducing energy use, contributing meaningful guidance for architects, planners, developers, and policymakers. By combining simulation-based research with field validation, this thesis deepens understanding of the complex relationship between urban form, thermal comfort, and energy performance, providing a foundation for resilient and socially sustainable neighborhoods in the face of climate change.Doctor of Philosophy (Ph.D.)2026-09-0
Investigating the Origin of Low-Mass Quiescent Galaxies
No full textA significant fraction of galaxies today have ceased forming appreciable amounts of new stars. In the past several decades, both theoretical and observational work have presented several mechanisms capable of driving this so-called "quenching". However, many of these mechanisms are primarily associated with a buildup of stellar mass, and thus cannot explain the increasing numbers of low-mass quiescent galaxies unveiled by current-generation observatories like the James Webb Space Telescope (JWST). In this thesis, I investigate these populations of low-mass quiescent galaxies and seek to explain both how they form early in the history of the Universe and their subsequent evolution over billions of years of cosmic time. First, I present the structural measurements of over 50,000 galaxies from the COSMOS-DASH survey. These measurements revealed a distinct change in the relation between galaxy size and stellar mass, which occurs at roughly 10^10 Msun. This flatter size-mass relation suggests that significantly different processes drive the formation, evolution, and quenching of low-mass galaxies relative to their massive counterparts. One mechanism that could explain low-mass quenching is gas removal driven by centrally concentrated starbursts. In the second part of this dissertation, I measure the star-formation histories of the central and outer regions of intermediate-mass star-forming galaxies at z~2.3, finding near-ubiquitous evidence of relatively recent (900 galaxies in 3D-DASH, I make preliminary conclusions on the different quenching pathways that affect massive galaxies. Specifically, I find a trend between global formation times and quenching timescales, implying that, in general, massive galaxies that form early quench slowly, while late forming sources quench rapidly. I also find connections between quenching timescales and stellar surface densities, suggesting that rapidly quenched galaxies tend to be more compact while slow quenchers are more extended.Doctor of Philosophy (Ph.D.
The Fleeting Story: Bringing Yoshimoto Banana’s “Utakata” Into the Spotlight
No full textYoshimoto Banana’s “Utakata,” published originally in 1988 as part of the two-novella compendium Utakata/Sankuchuarī, is one of her more overlooked stories. Though it was written mere months after the explosive debut of the first half of Kitchen, and though it did not sell much less than any of her other early works, the story has since been largely neglected by scholars in discussions of Yoshimoto, and it has remained untranslated in full until the present day. This is unexpected, as “Utakata” appears on the surface to share the same core themes with Kitchen as well as with one of her other highly popular novels, Tsugumi. However, a closer investigation reveals minute yet significant differences in the presentation of Yoshimoto’s usual themes within “Utakata.” In contrast to Kitchen and Tsugumi, which have received praise for challenging social norms and taboos—the former especially so—“Utakata” does not appear to present any particularly radical ideas or conclusions. This may constitute one of the key reasons that it never received much critical or international attention. Nonetheless, as a story, “Utakata” holds the potential to be quite valuable on multiple levels—personal, societal, and global—and as such, it is certainly worthy of a translation into English. Also discussed are the approaches and decisions used in translation, and a full English translation is included as a supplemental document.Master of Arts (M.A.
Chaotic lensed billiards
No full textLensed billiards are an extension of the notion of billiard dynamical systems obtained by adding a potential function of the form, were is a real-valued constant and is the indicator function of an open subset of the billiard table whose boundaries (of and the table) are piecewise smooth. Trajectories are polygonal lines that undergo either reflection or refraction at the boundary of depending on the angle of incidence. Our main focus is to explore how the dynamical properties of these models depend on the potential parameter using a number of families of examples. In particular, we explore numerically the Lyapunov exponents for these parametric families and highlight the more salient common properties that distinguish them from standard billiard systems. We further justify some of these properties by characterizing lensed billiards in terms of switching dynamics between two open (standard) billiard subsystems and obtaining mean values associated to orbit sojourn in each subsystem
Ecological Trait Differences Are Associated with Gene Expression in the Primary Visual Cortex of Primates
No full textPrimate species differ drastically from most other mammals in how they visually perceive their environments, which is particularly important for foraging, predator avoidance, and detection of social cues. Background/Objectives: Although it is well established that primates display diversity in color vision and various ecological specializations, it is not understood how visual system characteristics and ecological adaptations may be associated with gene expression levels within the primary visual cortex (V1). Methods: We performed RNA-Seq on V1 tissue samples from 28 individuals, representing 13 species of primates, including hominoids, cercopithecoids, and platyrrhines. We explored trait-dependent differential expression (DE) by contrasting species with differing visual system phenotypes and ecological traits. Results: Between 4–25% of genes were determined to be differentially expressed in primates that varied in type of color vision (trichromatic or polymorphic di/trichromatic), habitat use (arboreal or terrestrial), group size (large or small), and primary diet (frugivorous, folivorous, or omnivorous). Conclusions: Interestingly, our DE analyses revealed that humans and chimpanzees showed the most marked differences between any two species, even though they are only separated by 6–8 million years of independent evolution. These results show a combination of species-specific and trait-dependent differences in the evolution of gene expression in the primate visual cortex