UC San Diego

Cholera causes severe diarrheal illness in young children, but the mechanisms underlying age-dependent susceptibility remain unclear. Experimental cholera in neonatal mice recapitulates age-dependent susceptibility: suckling mice are susceptible to Vibrio cholerae colonization and cholera toxin (CT)-dependent disease, whereas adult mice are not readily colonized and do not develop cholera-like disease. Here, we define a developmental window in which susceptibility declines sharply over the first two postnatal weeks.To determine whether the gut microbiota contributes to this developmental transition, we first tested whether short-term microbiota manipulation could alter susceptibility to V. cholerae. Microbiota from both neonatal and adult donors protected adult germ-free mice from V. cholerae colonization, but adult microbiota transplants did not readily alter the neonatal gut microbiota or confer resistance to susceptible neonatal mice. In contrast, transplantation of intestinal contents from 14-day-old donors reduced disease severity and neonatal recipients, and sterile filtered 14-day-old small intestinal contents reduced both V. cholerae colonization and disease. These findings suggest that developmental microbiota-associated factors can rapidly influence susceptibility independently of stable donor microbiota engraftment.We next investigated how early life disruption of microbiota assembly affects the developmental emergence of resistance. Maternal antibiotic exposure disrupted vertical transmission of maternal microbiota to offspring and altered distal small intestinal microbiota assembly, extending the window of susceptibility to CT-dependent V. cholerae colonization and disease. Pups born to antibiotic-treated dams exhibited reduced Lactobacillaceae and increased Enterobacteriaceae, and reintroduction of an endogenous Lactobacillus isolate restored offspring lactobacilli levels and reestablished resistance to experimental cholera at two weeks of age. Consistent with a direct protective role, increasing lactobacilli in susceptible neonatal mice reduced experimental cholera burden, and cultures of the endogenous Lactobacillus isolate as well as spent culture media acidified the in vitro growth environment and rapidly eliminated recoverable V. cholerae. Together, these findings support a model in which the emergence of resistance to experimental cholera during early life is associated with maturation of the gut microbiota and acquisition of specific microbiota-associated protective functions. These findings also identify vertical transmission of maternal microbiota to offspring and lactobacilli-associated antagonism as determinants of early-life resistance to cholera.

The human brain is an organ of prediction. It does not passively receive sensory information but actively constructs perception, action, and the felt sense of the body by issuing precision-weighted predictions and updating them against ascending prediction error. Under this hierarchical predictive architecture, suffering and well-being are not only outcomes of peripheral physiology but configurations of the brain's generative model: how rigidly its priors are held, where its attention is allocated, and how its high-level beliefs are coupled to interoceptive and exteroceptive signals. This dissertation investigates whether, and how, mind-body techniques transiently reorganize the predictive hierarchy, and what neural, molecular, and computational signatures accompany that reorganization.Three complementary studies are presented. Chapter 1 reports a randomized controlled trial demonstrating that four days of mindfulness meditation training produces a 33% reduction in experimentally induced pain that is predicted, within meditators, by a meditation-induced rise in the Shannon entropy of cortical functional connectivity networks, consistent with a flattening of the prediction-driven attractor landscape that ordinarily sustains pain. Chapter 2 reports a multi-omics, fMRI, and behavioral characterization of a seven-day retreat combining meditation, cognitive reconceptualization, and open-label placebo healing rituals, showing that the intervention disengages the default mode and salience networks, increases circulating BDNF and endogenous opioid pathway proteins, enhances glycolytic metabolism, and modulates exosomal neurotransmission-related microRNAs. This coordinated brain-body reorganization is interpreted as the joint relaxation of high-level priors, the recalibration of embodied first priors, and the reopening of the predictive system to new beliefs. Finally, Chapter 3 develops a unifying theoretical framework arguing that serotonergic psychedelics, contemplative practice, and caloric restriction converge on a single computational endpoint, a transient near-critical regime of relaxed priors and expanded degrees of freedom in the generative model, and that the depth of the resulting mystical-like experience is the primary therapeutic agent. Together, these studies advance an empirically grounded, computationally explicit account of how transformative brain-body states are produced, how they heal, and how the science of such states may be pursued with greater precision.

Students with borderline intellectual functioning (BIF), typically defined as IQ scores between 70–85, occupy a particularly vulnerable and understudied position in secondary transition. They often do not qualify for intensive intellectual disability services yet struggle to meet traditional academic and graduation expectations, leaving them underserved by existing structures. This qualitative multiple case study design, incorporating cross-case analysis, examined the transition experiences of seven high school graduates with BIF from a comprehensive high school in Southern California, guided by Self-Determination Theory. Data included document analysis of IEPs, ITPs, and transcripts, alongside semi-structured interviews. Cross-case thematic analysis revealed four interconnected findings: self-determination thrived when students felt competent and heard; relationships and interest-based learning were powerful yet underutilized; life-skills instruction was fragmented; and connections to external supports were weak. These findings suggest that identical graduation pathways do not produce equity for students with BIF. The study offers implications for IEP case managers, school and district leaders, and recommendations for more flexible pathways, robust life-skills preparation, and transition practices centered on student voice and self-determination.

Soft materials and structures are increasingly important in biomedical systems, wearable technologies, soft robotics, and architected materials because their compliance and capacity for large deformation enable functions difficult to achieve with conventional rigid materials. Their reliable application, however, requires predicting when deformation localizes and fracture compromises structural integrity. This becomes challenging in systems involving nonlinear deformation, complex geometry, anisotropy, discrete structural organization, or time-dependent material response. This dissertation develops physically interpretable fracture-prediction frameworks based on continuum mechanics, using analytical and computational models to capture the governing relationships among loading, geometry, constitutive behavior, and fracture.The work begins by investigating whether continuum fracture mechanics can provide a useful predictive framework for a materially continuous but microstructurally complex soft system. Aortic dissection, a life-threatening condition whose defining mechanical process is tear propagation within the aortic wall, provides a direct and compelling application of fracture mechanics. By representing the wall as an effective continuum, a semi-analytical model relates loading, geometry, and constitutive behavior to the fracture-driving force. Comparisons with finite-element calculations and reported aortic fracture resistance demonstrate how the framework can identify mechanically favorable conditions for tear propagation. The dissertation then examines whether continuum fracture concepts remain applicable to periodic soft lattices. Although homogenization provides a simpler description than full-scale models that explicitly resolve individual unit cells, its connection to local rupture is not immediately evident. For elastic hexagonal lattices, the macroscopic driving force for fracture is related to the local deformation at the crack tip, enabling fracture conditions and structural trends to be identified with reduced computational effort. This continuum-to-lattice framework is subsequently extended to viscoelastic lattices, where crack evolution is time-dependent even under fixed macroscopic deformation. Rather than prescribing an independent crack-growth or traction law, a local failure criterion is combined with physically motivated assumptions and event-resolved finite-element calculations. The resulting reduced-order fracture model evaluates how viscoelasticity affects delayed failure, intervals between successive events, and overall crack-evolution trends. Together, these studies demonstrate how continuum mechanics-based fracture models can be formulated and extended as soft systems progress from effective continuous materials to discrete nonlinear structures and, finally, to time-dependent viscoelastic fracture.

This dissertation examines the uneasy fit between geography and racialized, gendered laboring bodies across Global Asias, with a focus on Hong Kong, Singapore, and mainland China. Rather than treating Asia as a flat territorial surface, I approach it as a 3-D, volumetric formation expanding across land, water, air, and digital axes. This geo-technological expansion depends on migrant and diasporic laboring bodies, whose varied forms of work shape how they move through, inhabit, and sense the landscape. Yet they are rendered a sliding anomaly: indispensable to urban ideals of hygiene, order, and progress, but readily recast as disorderly, contagious, or threatening, their position shifting across this broad spectrum as conditions demand. I theorize this unstable body–space relation as a geo-somatic misfit, in which bodies and landscapes alternately align and misalign, revealing multiple sensory, affective, and material relations to place.Drawing on urban planning maps, literature, film, and visual art, I trace laboring bodies and body–space misfits from the 1970s into contemporary and speculative futures. Across three chapters, I move through interconnected sites in Global Asias. Chapter 1 examines Hong Kong’s urban mythology of gong tau, or Southeast Asian sorcery, and the rise of erotic hex cinema from the 1970s to the 2000s. I situate this occult imaginary within Hong Kong’s changing architecture and cross-border ethnosexual economies, tracing a geospatial morphology that binds migrant sex workers to evolving building forms and public health discourses of disease. Chapter 2 turns to Singapore’s reclaimed landscapes and migrant construction workers’ remapping practices as forms of counter-volume. Chapter explores speculative Asian geographies, from a fused robotic Asia to expanding cyberspace, and the multiplication of labor across organic, synthetic, and digital bodies. These sites are not isolated dots but nodes within broader misfit geographies, where laboring bodies expose the uneasy fit between local landscapes and Global Asia.

Lithium-ion batteries anchor transportation electrification and grid-scale energy storage. Usable capacity fades with age. Safe operation, second-life screening, and investment decisions therefore largely depend on the capacity-based state of health (SoH). A reference capacity test measures it directly but is costly and interrupts service. This dissertation connects health estimation to economic value in three parts.The first part estimates SoH for cells cycled under randomized loads mirroring real-world usage. Each cycling stretch is summarized by cell voltage and temperature probability density functions. The expectation–maximization algorithm identifies the assumed bimodal voltage and normal temperature distribution parameters. Identified parameters and statistical descriptors most correlated with SoH feed a support vector regression. On the NASA random walk benchmark it reaches 2.63 % mean absolute error and 3.31 % root mean square error. Its maximum error, 6.77 %, is the lowest of forty algorithm-input combinations. Moreover, the accuracy extends to three more predictable test groups. The second part carries SoH estimation to three retired electric-vehicle packs. These packs supply scarce labels and sense only voltage and current. A Transformer encoder is pretrained as a masked autoencoder on raw cycling data and fine-tuned on labeled stretches. A transfer study shows that corpus composition, not size, decides whether pretraining helps. A pool mixing a wide-SoH public NASA dataset and the second-life packs lowers error on the two lithium-manganese-oxide packs, which make up most of this pool, by up to 35.5 % versus no pretraining. On the chemistry-distinct nickel-cobalt-aluminum-oxide pack, however, the same pool raises error by 41.6 %. The NASA dataset alone is instead the most accurate and most stable source for that pack. The third part turns from battery health to economic worth. A dispatch model maximizes energy-trading revenue subject to a battery model’s constraints. Piecewise linear approximation and McCormick envelopes relax them into a mixed-integer linear program. Vanadium redox flow, sodium-ion, and lithium-ion systems, dispatched for energy time shifting against 2022 California real-time prices, earn aggregated daily revenues of $111.08, $67.63, and $31.17. Higher C-rate and stronger power capability let a system cycle more often and earn more per cycle.

The sensors required to mimic clinical metrics of anything from pupil change to blood pressure to functional strength are already present in the smartphones that billions of people carry in their pockets. The barrier to using them for clinical measurement is not hardware, it is software. This dissertation presents research developing and validating smartphone-based sensing systems across three clinical domains: neurological assessment, cardiovascular monitoring, and musculoskeletal assessment, with two paired goals: expanding access to established clinical measurements and facilitating the discovery of novel digital biomarkers. In neurological assessment, two smartphone pupillometry systems are presented, one repurposing the near-infrared facial recognition camera for absolute pupil diameter measurement (median MAE 0.27 mm against a clinical pupillometer) and one using far-red optical filtering for universal RGB camera compatibility (mean Pearson correlation 0.93, MAE 2.4%), with both systems applied toward mild cognitive impairment screening via pupillary response features. In cardiovascular monitoring, the first absolute oscillometric blood pressure measurement on an unmodified smartphone is achieved through Vibrometric Force Estimation (VFE), which uses the built-in vibration motor and inertial measurement unit to measure fingertip force without dedicated hardware, yielding systolic MAE of 9.21 mmHg and AUC of 0.92 for hypertension screening. In musculoskeletal assessment, inverse kinematics features extracted from inertial measurement unit data during the 30-second chair stand test substantially outperform the conventional repetition count for body composition prediction, achieving Pearson correlations of 0.94 for total lean mass and 0.73 for percent body fat against dual-energy X-ray absorptiometry in 80 participants. Across all three domains, systems are anchored to universally available smartphone sensors, validated against clinical gold standards, and evaluated for demographic equity, demonstrating that clinically relevant measurement and biomarker discovery are achievable through software repurposing of hardware that already exists in consumers' hands.

Spinoza's monistic metaphysical picture is unique in its uncompromising and radical commitments. According to Spinoza, there is one substance—God—of which everything else (trees, humans, and chemical elements) are modes. Thus, all things are causally and conceptually dependent on God, inhere in God, and are effects of God's infinite essence, from which everything necessarily follows. Spinoza is also a committed necessitarian and determinist, according to whom God alone is a free cause, acting, as do all things in nature, from the universal laws of its own nature. According to this view, all finite things are on a metaphysical par, existing as parts of an infinite chain of finite causes as finite and determinate expressions of God’s infinite power.These austere metaphysical commitments have led some readers to question the reality of anything but God and to adopt an acosmist view. Furthermore, according to Spinoza's epistemology, our experience of singular things yields only partial and confused knowledge. His definitions of both individuals and singular things, in turn, cast doubt not only on the existence of finite, singular entities but also on their particular identity and metaphysical stability. In other words, Spinoza's philosophy as a whole does not seem to align with our common-sense understanding of the world and the things within it. While it is relatively clear what substance, or Natura Naturans, is for Spinoza, it is less clear what he takes finite things to be exactly, or what Natura Naturata consists of.This dissertation delivers an account of the metaphysics of the finite in Spinoza, focusing mainly on his Ethics. I argue that singular things are relationally constituted and exist only durationally, and that the identity of each and every singular thing is thus relational. I further argue that singular things' identity is grounded solely in their actual essence, i.e., the conatus, which I understand as a strive to increase in power, necessarily and indispensably involving interactions with others. Thus, I argue that when singular things cooperate with things of the same nature, their identities are genuinely metaphysically united. This dissertation further argues that Spinoza's concept of an individual, defined by a fixed ratio of motion and rest, determines only a thing's kind, yet not its particular identity. It also argues that formal essences, understood as mediate infinite modes and thus as "concrete universals," express such ratios of motion and rest and are shared kind-essences.

In the past few years, the landscape of software development has been reshaped. Large Language Models have made generating code, proofs, tests, and documentation cheap and abundant. A generated artifact may look plausible yet be subtly wrong. At the pace generation now runs, manual review cannot close the gap. Generation is cheap; review is the bottleneck. This dissertation investigates how to establish formal guarantees for code at scale, with minimal human oversight. The first part of this dissertation studies mechanisms for enforcing strict syntactic constraints during LLM generation. In particular, we focus on Grammar-Constrained Decoding (GCD), which integrates formal grammar parsers directly into the decoding process. To make this approach practical for modern LLMs with large vocabularies, we develop techniques to improve the efficiency of grammar-aware decoding. We also identify a fundamental theoretical limitation of GCD: although it guarantees syntactic validity, it can distort the model’s learned probability distribution. To address this limitation, we introduce the Grammar-Aligned Decoding (GAD) problem, which seeks to align the sampling distribution with the desired structural constraints. While syntax can often be captured by explicit grammars, semantic correctness is substantially harder to specify. The second part of this dissertation addresses the problem of synthesizing precise semantic specifications from source code. We introduce a counterexample-guided synthesis algorithm that generates sound and precise program specifications within a target domain-specific language. We further extend this algorithm to nondeterministic programs, enabling the synthesis of both over-approximate and under-approximate specifications. The third part of this dissertation builds on these component-level specifications to synthesize complex systems compositionally. Rather than verifying or generating a large system monolithically, our approach constructs larger programs from existing sub-programs and verified lemmas. Each synthesis task is reduced to a smaller subproblem of the same general form, keeping the search space and proof obligations bounded at the component level. Together, these techniques provide a robust path toward the scalable, specification-guided construction of verified software systems.

Time series data often exhibit high dimensionality, temporal dependence, distributional heterogeneity, and limited labeled observations. These features pose substantial challenges for statistical estimation, prediction, and inference. This dissertation develops statistical methods for dependent data, with particular emphasis on high-dimensional regression, discrete-valued time series, predictive inference, and transfer learning. First, a factor-augmented regression framework is developed for high-dimensional time series. By separating pervasive cross-sectional and temporal dependence from sparse predictive signals, the proposed method improves estimation and variable selection when the observed covariates exhibit strong dependence. Second, statistical inference for high-dimensional binary time series is challenging because the underlying transition structure becomes difficult to estimate as the dimension grows. A generalized binary vector autoregressive model is developed to characterize dynamic dependence among binary-valued processes. Post-selection inference procedures are further established using Gaussian approximation and wild bootstrap methods, providing valid inference in high-dimensional time series settings. Moreover, conformal prediction provides model-free predictive inference under exchangeability or independence assumptions. We develop distributional conformal prediction for dependent observations and establish theoretical coverage guarantees for its application to Markov processes. The proposed framework exploits the underlying temporal dependence while retaining rigorous uncertainty quantification. Finally, estimation can be difficult when only a limited number of observations are available in the target time series, while directly pooling data from different sources may introduce bias due to distributional heterogeneity. We extend a transfer learning framework to high-dimensional time series, allowing information to be integrated from similar and related time series. The resulting methods improve estimation and prediction when target-domain observations are limited, with theoretical guarantees.