71446 research outputs found
Sort by
Advancing Deep Learning for Multiagent AI: Mechanisms, Organizations, and Dynamics
No full textEffective and safe multi-agent AI holds the promise to solve important problems across domains such as digital markets, multi-robot coordination, human-AI alignment, and modular methods for monolithic learning systems.
This thesis aims to advance deep learning methods so that they scale to and provide certifiable alignment guarantees in real-world complexity. Technically, it approaches this goal by encoding structure into deep learning that mediates multi-agent dynamics and enables rigorous analysis. This agenda explores systems of independent agents as well as embodied systems, modeled compositionally as multiple agents. This approach has produced a series of deep learning methods that are the first of their kind, achieving provably aligned and effective multi-agent AI that synergizes insights from computational economics, game theory, reinforcement learning, and robot learning.
Specifically, this agenda operates at three macro-to-micro levels of system abstraction, focusing on incentive structures that shape learning outcomes, interaction structures that simplify dynamics, and statistical structures that guide agent learning, respectively. Key contributions are as follows:
Incentive structure: At this level, I design principled incentives that make a target behavior an equilibrium to which rational agents converge, avoiding the need to analyze dynamics in detail. To achieve this, I encode a menu-based architecture and function discontinuity into deep learning to address landmark problems in incentive design, including auction and contract theory. Applied to computational economics, this method yields the first provably aligned (truthful reporting) and effective (revenue optimizing) neural mechanism for multi-bidder, multi-item auctions and single-bidder combinatorial auctions.
Interaction structure: At this level, I accelerate learning by encoding structures that mediate agent interaction, such as groupings, symmetry, and near decomposability into deep learning. These structures simplify multi-agent dynamics and boost learning efficiency in the control and design of a single embodied agent, and also promote multi-agent cooperation.
Statistical structure: At this level, I identify statistics that characterize the dynamics of a system. The problem of agent learning is formulated as modeling and thereby steering these dynamics. I propose a unified approach for dynamics modeling using a generative neural ordinary differential equation that encodes the dependency structure among the statistics. I also study agent modeling using large language models, with the aim of optimizing decision-making policies based on these models. These studies are supported by principled analyses that rigorously assesses the effectiveness of dynamics modeling and policy learning.Engineering and Applied Sciences - Computer Scienc
Statistical Estimation of Circadian Time Using Gene Expression Data
No full textThis dissertation develops statistical methods and evaluation standards for estimating circadian time using high-throughput gene expression data. The first paper introduces an alternating weighted least squares framework that robustly infers circadian time, with strong generalization across tissues and platforms. The second paper extends this framework by incorporating batch adjustment through low-rank latent factors, yielding accurate performance on both simulated and real batch-confounded data. The third paper introduces a unified benchmarking framework that standardizes preprocessing, choice of assessment, and performance metrics, and evaluates a comprehensive selection of circadian time estimation algorithms across diverse contexts. Together, these studies both establish a comprehensive benchmarking approach for circadian time inference and demonstrate a modeling strategy that is robust, transferable, and state-of-the-art across real and simulated settings.Biostatistic
Click Here to Verify You’re Human
No full textThis short fiction collection offers a cultural critique of social technology, social judgment, and social exchange. From a content creator who uses AI to misrepresent her entire existence for profit, to drunk suburban moms that fist fight over a bingo prize at a school fundraiser, the characters reveal their deeply human authenticity, even when their behavior is inhumane. The stories blend humor with subtextual sadness, leaving the reader laughing, wondering how we got here, and a little bit scared of where we go next.Extension Studie
Belinda Sutton's Appeal: Slavery, Political Activism, and Legal Claims in Revolutionary New England
No full textThis dissertation tells the story of an enslaved mother’s appeal to Revolutionary America. Following the shadowy trail that leads from the life Sutton lived to the petition she authored, it provides an intimate sense of Sutton as a historical actor in her own right, as a mother who once walked down Boston’s cobbled streets. The trail leads to the rich political and intellectual world of Black New Englanders. Sutton was thoroughly embedded in this world, a milieu out of which sprang the first antislavery organization in the British empire—a committee of enslaved and formerly enslaved people whose antislavery campaign abolished slavery in Massachusetts.
Sutton’s petition came on the heels of that campaign. Uncovering new details about the effort, I track its activists in the courts and the legislature, in the famous town meetings and Committees of Correspondence, in the press and the streets. Doing so reveals that Sutton drew on, and contributed to, a rich tradition of Black legal and political advocacy. It reveals, too, that Belinda Sutton and the Africans in her world had long made claims about what the law is or should be, about their rights, about what actions merited legal remedy. Together, they formulated their own critiques of slavery that shaped local legal culture and created new legal norms.African and African American Studie
A History Written in Flesh: The Embodied Archive of the Sixteenth-Century Tupinambá
No full textIn this dissertation, I fundamentally re-examine the historiography of sixteenth-century Brazil by challenging the colonial construction of the Indigenous Tupinambá as an "impossible subject" — a people deemed ahistorical or inherently unintelligible within Eurocentric frameworks. I propose and enact a methodology for reading the Tupinambá body itself as a sophisticated "corporeal archive" of historical and relational memory, moving beyond traditional scholarly reliance on fragmented colonial textual archives. Through a critical analysis of Portuguese and French colonial chronicles, particularly the writings of Gabriel Soares de Sousa, I deconstruct the colonial gaze that distorted Indigenous realities while simultaneously uncovering the subtle traces of Tupinambá agency and complex social systems. I explore how European observers interpreted male bodies, practices of warfare, and cannibalism, as well as the profound yet often misunderstood roles of Tupinambá women in rituals of weaving, wounding, and weeping. These analyses demonstrate how Indigenous bodies were not merely subjects of colonial description, but active sites where kinship, memory, and collective history were continuously inscribed and performed. For example, I show how warrior scars and women's ritualistic body modifications served as complex systems of communication, marking personal achievements, collective grief, and intergenerational ties, effectively encoding a history distinct from European textual forms. Ultimately, I argue that the history of the Tupinambá was never absent, but rather written in a different medium — one composed of bodily transformation, ritual, and lived experience rather than ink and paper. By treating bodily inscriptions and practices as a coherent system of memory, this research not only offers a nuanced reconstruction of Tupinambá society but also forwards a significant methodological contribution that expands the very definition of the archive. It provides a crucial tool for listening to other historically silenced peoples whose own records may be similarly embodied and relational, challenging the colonial monopoly on what constitutes history itself. In tracing these embodied practices, I contribute to histories of Indigenous studies, colonial Brazil, and the theoretical understanding of archives and memory.Romance Languages and Literature
EEG-Based Biomarkers of Neural Circuit Function in Rett Syndrome
No full textAtypical sensory processing is a key feature of many neurodevelopmental disorders, likely contributing to higher-order deficits in cognition, communication, and behavior. However, the neural mechanisms underlying these abnormalities remain incompletely understood. Rett syndrome (RTT) is one neurodevelopmental disorder, related to autism spectrum disorder and predominantly caused by de novo mutations in the X-linked MECP2 gene. RTT is characterized by an initial period of typical early development followed by rapid regression leading to profound motor, communication, autonomic, and other impairments; disruptions to cognitive function and sensory processing have more recently emerged as core features. Electroencephalography (EEG) at rest, as well as time-locked to sensory stimuli (event-related potentials, or ERPs), offers powerful tools to study the neural dynamics underlying these disruptions, with the potential to bridge neural circuit dysfunction with clinically relevant neurophysiological markers and behavioral outcomes. This dissertation integrates findings from three EEG-based studies in Rett syndrome. First, we aimed to investigate variability in neural responses to sensory stimuli by measuring inter-trial phase coherence (ITPC). We found increased trial-by-trial variability in individuals with RTT in response to both visual and auditory stimuli, correlating with dampened ERPs and atypical sensory processing phenotype. In the second study, we utilized resting-state EEG to investigate thalamocortical contributions to sensory processing dysfunction by measuring transient beta events (short bursts of beta-frequency activity linked to thalamocortical drive and implicated in tactile processing). We observed elevated transient beta event rates in RTT as well as other related neurodevelopmental disorders, pointing to disruptions within thalamocortical sensory pathways. Finally, we investigated interactions between oscillatory frequencies at rest in RTT by analyzing phase-amplitude coupling, a mechanism of neural coordination that supports a range of cognitive and sensory processes. Results demonstrated significantly increased theta-gamma and alpha-gamma coupling in RTT, and computational modeling further identified dysfunction in VIP+ interneurons as a candidate cellular mechanism underlying this enhanced coupling. Collectively, these findings advance the field’s understanding of the neural dynamics underlying altered sensory and cognitive processing in RTT, paving the way for use as functional biomarkers and providing insights that can inform future work.Neuroscienc
Developing and testing coarse-grained models of mitochondrial form and function
No full textMitochondria are multifunctional and semiautonomous organelles. They perform respiration in order to produce adenosine triphosphate (ATP), the energy currency which fuels cellular processes. However, the factors that control the rate of mitochondrial respiration and determine cellular mitochondrial content are incompletely understood. Working with the budding yeast Saccharomyces cerevisiae, we show that respiration rate is saturated and that mitochondrial biogenesis is decoupled from the production of other cellular components. Respiration rate is robust to acute perturbations of cellular ATP demand and nutrient supply. Instead, respiration rate is determined by mitochondrial amount due to saturation of the electron transport chain. Mitochondrial accumulation rate is robust to changes in growth conditions, resulting in mitochondrial amount being largely determined by cell division time. A long-standing observation is that in fast-growing cells, respiration rate declines with increasing growth rate and is compensated by an increase in fermentation, even though respiration is more efficient at producing ATP -- a phenomenon known as aerobic glycolysis. Our model of respiration and mitochondrial growth, combined with a kinetic model of glucose uptake, provides a quantitative explanation for aerobic glycolysis in budding yeast.Biology, Molecular and Cellula
Nervous system function and behavior in C. elegans embryos
No full textEvents in early brain development are known to have profound impacts on later-
life developmental outcomes. Most studies of circuit function and behavior use post-
natal samples, likely due to the inaccessibility of intact embryos. How does nervous
system function emerge and begin to control behavior during embryonic development?
The optically accessible C. elegans embryo provides a powerful tool with which to
address this question.
GABA signaling is proposed to play important roles in circuit formation and
plasticity, making GABA an interesting molecule to study in the context of embryonic
nervous system function. We find that GABA release from developing motor neurons
occurs prior to localization of pre- and post-synaptic ion channels in the nerve ring.
Synaptic vesicle release contributes to but is not required for early motor neuron GABA
release. Although bestrophin channels are known to mediate non-synaptic GABA
release from glia, we find that bestrophins act in neurons to promote synaptic GABA
release. These results indicate a role for non-synaptic GABA transmission in prenatal
circuits.
Previous work described a sleep-like behavior in late embryonic development.
We took two approaches to identify developmental and circuit mechanisms underlying
late embryonic behavior. First, we screened genes associated with autism spectrum
disorder (ASD), which is theorized to have roots in early neurodevelopment. Several
ASD risk genes regulate embryonic sleep-like behavior. An allele of CaV1 (with an
equivalent amino acid change to a Timothy Syndrome allele) alters embryonic behavior
by its effects downstream of the sleep active neuron RIS. Additionally, genes required
for sensory neuron function are required for wild type behavior, suggesting sensory
neurons regulate circuit function before the presence of externally generated sensory
cues. These results define elements underlying embryonic neural circuit function.Biological and Biomedical Science
