49755 research outputs found
Sort by
Barriers and Breakthroughs: Business Models for Electric Motorcycle Taxis in Bangkok
No full textThe transition to electric mobility offers a promising pathway for sustainable urban transport, yet adoption in informal sectors remains poorly understood. This paper examines four business models for promoting electric motorcycle uptake among Bangkok’s motorcycle taxi drivers: full ownership, partial battery leasing, full battery leasing, and rental. Adopting a case study, the research draws on stakeholder interviews and pilot project documentation to identify systemic barriers and enabling factors.
Three system-level constraints limit scalability: network effects in which inadequate and brand-specific battery-swapping infrastructure discourages adoption and investment; sociotechnical misalignments between new business approaches and entrenched regulations and driver routine; and justice-related exclusions which disadvantage informal, low-income operators through financial risks and policy neglect. Finding reveals that high upfront costs, income volatility, resale uncertainty, and restrictive registration rules intersect with infrastructure fragmentation to stall adoption despite pilots demonstrating feasibility.
Policy recommendations include reforming vehicle registration to allow fleet-based ownership, expanding and standardizing battery-swapping networks, introducing flexible financing and resale guarantees tailored to informal workers, and co-designing programs with drivers to align with occupational culture and needs. By reframing adoption challenges through network, institutional, and justice lenses, the paper offers insights for scaling low-carbon mobility transitions in informal transport sectors across the Global South.
Keywords: Motorcycle taxis, Electric motorcycles, EV adoption, Business models, Informal transport, Global South
Breaking the Mold: Business Model Innovation in Entrepreneurship Education
No full textEntrepreneurship education within community colleges is at a critical juncture, as technological disruption, shifting demographics, and alternative credential providers challenge established academic paradigms. The research informing this dissertation seeks to answer an urgent, broad corollary question: “How do community colleges maintain relevance and competitive advantage in a rapidly evolving landscape of entrepreneurship education?” It probes this question by investigating how three San Francisco Bay Area community colleges can harness business model innovation to strengthen their entrepreneurship programs, enhance student success, and maintain competitiveness in an evolving market. Grounded in the Resource-Based View (RBV) and Strategic Leverage Through Learning (SLL) framework, the study highlights the interplay of tangible resources (e.g., location, funding, facilities) and intangible capabilities (e.g., faculty expertise, industry partnerships) as engines of sustained advantage.
Drawing on a qualitative multiple-case study methodology, the research illuminates how each college strategically aligns curricular design, co-curricular offerings, and external collaborations to amplify practical entrepreneurship training. Semi-structured interviews, surveys, and field observations expose the organizational silos and regulatory hurdles that often limit innovation while also revealing the rich potential for interdepartmental synergy and bold partnerships with local incubators, tech companies, and community stakeholders. By tracing how faculty, administrators, and industry partners co-create experiential learning opportunities—from industry and student-led workshop series, skills-based projects, rapid prototyping, pitch and demo competitions, and pop-up markets to startup accelerators—the dissertation provides a nuanced view of how an entrepreneurial ecosystem can meaningfully bridge theory and practice.
The findings underscore the crucial role of organizational learning in fostering innovation. SLL provides a powerful mechanism by which colleges can align learning processes with institutional goals, allowing them to adapt quickly, share knowledge effectively, and co-create value with stakeholders. This learning model transforms isolated initiatives into system-level innovations, enabling institutions to respond proactively to disruption. Key enablers of success include cross-disciplinary collaboration, inclusive skill-building, robust technological infrastructure, and ongoing faculty development.
This dissertation proposes an open innovation platform grounded in strategic organizational learning, supported by a viable business model for the future of community college entrepreneurship education. By embedding SLL at the core of the open innovation platform, colleges can foster boundary-spanning collaborations for resource sharing, leverage knowledge flows between organizations, academia, students, and other stakeholders, and support experiential learning practices. In doing so, entrepreneurship education is repositioned not merely as an academic offering but as a necessary life skill for socio-economic mobility and institutional strategy for advancing equity, relevance, and student success. The dissertation calls on policymakers, administrators, and scholars to reimagine community college entrepreneurship education as a dynamic, network-oriented enterprise equipped for the complexities of the 21st century
Self-Supervised Learning for Self-Modeling Robots
No full textHumans form internal representations of their bodies and capabilities through constant interaction with their environment. This ability motivates the creation of autonomous robots that can learn and adapt similarly. The central question driving this dissertation is how to equip robots with comparable, self-contained learning processes that unfold without continuing human supervision.
To address this need for human-like adaptability, we propose a framework consisting of two core methodologies: (1) Self-modeling, where deep neural networks capture a robot’s intrinsic attributes, from morphology, kinematics, and dynamics to human-like cognition. (2) Self-supervised learning, in which the robot’s own sensory streams furnish the training data, thus removing the reliance on manually annotated datasets. By combining these methods, our approach lays the groundwork for robots to evolve their knowledge and skills over time, mirroring the fluid developmental progression observed in humans.
Experimental results demonstrate that robots equipped with self-models and self-supervised learning capabilities exhibit higher resilience, faster adaptation, and greater autonomy, even in the face of morphological changes or hardware malfunctions. These findings conclude that combining self-modeling with self-supervised learning establishes a promising foundation for lifelong robotic learning, paving the way for highly robust, improving autonomous systems capable of operating in unstructured environments
High-Voltage and Adaptive Digital Power Management IC
No full textThe rapid development of power consumption across data centers and automotive applications has catalyzed significant advancements in power delivery system architectures. Single-stage high-voltage DC-DC converters are emerging as a promising alternative to traditional two-stage designs, offering reduced power losses and enhanced efficiency. However, achieving high efficiency across wide load ranges while ensuring long-term reliability poses several challenges.
First, the high voltage stress and large conversion ratios introduced by high-voltage inputs exceed the capabilities of conventional half-bridge topologies and standard MOSFET technology. Addressing this requires innovative power stage topologies and advanced materials or technologies for power switches. Furthermore, the wide load current variability, influenced by the workload of the loaded chip, demands that converters maintain high efficiency across diverse operating conditions. Efficiency tracking techniques are essential to balance power loss components and optimize performance under varying loads. Reliability is another critical concern due to the high voltage, elevated temperatures, extended operational periods, and heavy load conditions typical of high-voltage DC-DC converters. Health monitoring circuits and control mechanisms are necessary to sense and mitigate potential risks. Additionally, most high-voltage DC-DC converters provide a single output voltage, while the connected chips often require multiple independent supply domains to optimize speed and power consumption. This calls for on-chip power management solutions, such as digital LDOs.
Chapters 2 through 5 of this thesis detail advancements in adaptive digital control to address these challenges. The first study presents a 24V-to-1V multi-level series-capacitor DC-DC converter with fast in-situ efficiency tracking and power FET code roaming, achieving up to 34.74% efficiency improvement and mitigating ON-resistance degradation by 5.8×. The second study introduces an online reliability estimation technique for ceramic capacitors, reducing estimation error by 11.4× compared to conventional offline methods. The third study demonstrates a GaN-based DC-DC converter with unified reliability and efficiency adaptive control, enhancing efficiency by 7.7% and reducing the threshold voltage degradation by 3.8×. Finally, the fourth study reviews advancements in digital LDOs over the past decade, proposes a new figure of merit, and identifies designs achieving superior transient performance
Fundamental Insights into Directed Filler Dispersion & Nanoplastic Formation in Semicrystalline Polymers
No full textFrom the start of the synthetic plastic era in the early 20th century to today, where over 400 million tonnes of plastic waste are produced each year, plastics have truly become one of the most dominant materials in our society in a relatively short amount of time. For materials that are still in their effective infancy, humans have made incredible progress in engineering polymers for everything from disposable cutlery to biocompatible medical implants. Yet despite this progress, there is still much more ground to cover. This thesis explores two main avenues of interest in polymer nanoscience, focusing specifically on semicrystalline polymers which comprise over two-thirds of polymer production: i) understanding and utilizing a method of directed filler dispersion and its impact on polymer composite properties and ii) degradation of semicrystalline polymers into micro- and nanoplastic.
Semicrystalline polymer composites and blends have been the subject of interest for the last several decades to tailor polymer properties and morphology towards specific applications. In chapters 2-4 of this thesis, we develop a fundamental understanding of crystallization-induced ordering of semicrystalline polymers. Crystallization-induced ordering results from the interplay between the semicrystalline polymer crystal growth rate and the diffusion of the nanofiller. This ordering process impacts the dispersion of the nanofiller resulting in assembly at different potential length scales (interlamellar, interspherulitic, etc.).
In chapter 2 we use poly(ethylene oxide)/silica nanocomposites to explore the impact of manipulating matrix molecular weight (which impacts matrix viscosity and therefore nanoparticle diffusion) on the crystallization-induced ordering of the nanoparticles. Through the use of small angle x-ray scattering and the Hermans orientation function we find that ordering in these systems can be divided into 3 regimes: i) when the crystal growth rate is much faster than nanoparticle diffusion, no change to dispersion occurs ii) when crystal growth rate and nanoparticle diffusion are balanced the strongest interlamellar ordering behavior is observed iii) when nanoparticle diffusion is much faster than the crystal growth rate then nanoparticles accumulate at the crystal growth front (interspherulitic).
Chapter 3 further develops our understanding by quantifying the diffusion of nanoparticles in the respective poly(ethylene oxide) matrices by using x-ray photon correlation spectroscopy (XPCS). Quantification of nanoparticle diffusion allowed for calculation of a dimensionless Peclet number between crystal growth rate and nanoparticle diffusion which showed that the highest interlamellar ordering was achieved when the Peclet number was of order unity. In addition, the dynamic XPCS measurements showed that the nanoparticle diffusion follows Stokes-Einstein predictions if a bound polymer layer (that scales with the polymer R_g) is considered in the size of nanoparticle.
Using the understanding gained from chapters 2 and 3, in chapter 4, we combine crystallization-induced ordering with a direction crystallization method called zone annealing to investigate the mechanical properties of miscible polymer blends with at least one crystallizing component. We use a model system of poly(ethylene oxide) (PEO) and poly(methyl methacrylate) (PMMA) due to their favorable miscibility in the melt state. Utilizing a slow crystallization rate allows for ordering of the amorphous PMMA in the interlamellar region of the semicrystalline morphology. Furthermore, we use zone annealing to orient the morphology unidirectionally resulting in anisotropic composite materials with an alternating nanostructure. We find that zone annealing had the most impact on the neat PEO mechanical properties, while modest property improvements were observed in the blends of PEO/PMMA. We found that having a glassy polymer (PMMA) in the interlamellar regions leads to a loss of an anisotropic property response, potentially due to an erasure of the property gradient between the crystalline lamella and amorphous interlamellar regions. Zone annealed blends of PEO/PMMA did show improved toughness compared to unorganized samples suggesting that properties of blends can be changed and improved using processing techniques to tune the composite nanostructure, which may serve as a potential way to upcycle polymer blends with one semicrystalline component.
Following the development of our fundamental understanding of crystallization-induced ordering, we investigate a new research area in nanoscience focused on the environmental degradation of plastic into micro- and nanoplastic. This research stems from the large amount of plastic waste in our environments that is subject to both mechanical and chemical degradation. Growing concerns over the fate of environmental plastic waste and the worrying potential health effects of nanoplastic accumulation have driven us to better understand these materials. While much of the research in this field has been focused on downstream effects of nanoplastic pollution (potential health effects, environmental identification, etc.) we focus primarily on the upstream processes that lead to the formation of nanoplastic. Gaining insights into the formation mechanisms of nanoplastic can allow us to develop better model systems for studies and potentially reveal ways to engineer plastics that reduce nanoplastic production. In chapter 5 we examine the formation and properties of poly(ethylene terephthalate) (PET) nanoplastic under quiescent accelerated environmental weathering conditions (hydrolysis) and correlate the degradation-induced embrittlement to the release of nanoplastic. In chapter 6 we begin to investigate the formation of PET nanoplastic through mechanical wear and compare these results to those from chemical degradation (hydrolysis).
In summary, we have used a suite of experimental tools to investigate the structure and properties that arise from crystallization-induced ordering in addition to understanding the formation of nanoplastic through different degradation mechanisms. As polymer production continues to increase, we will have higher performance polymeric materials, but we will also have additional plastic waste. Both topics covered in this thesis will continue to be relevant for the future development of polymeric materials. In the final future work section, we discuss further areas of research for all the sections covered in this thesis and propose a potential way for the two threads in this thesis to work together cooperatively. The fundamental polymer physics insights gained from these works lay the foundation to pursue and understand new routes to improve polymer performance while also looking to address the long-standing issues with environmental polymer waste that have come with the dominance of polymeric materials
Machine-Learning Fairness in Data Markets: Challenges and Opportunities
Full text linkMachine learning promises to unlock troves of economic value. As advanced machine-learning techniques proliferate, they raise acute fairness concerns. These concerns must be addressed in order for the economic surpluses and externalities generated by machine learning to benefit society equitably. In this thesis, we focus on the economic context of data markets and theoretically study the impacts of intervening to achieve machine-learning fairness.
We find that to effectively and efficiently intervene requires taking the data market into account in the design and application of the fairness intervention, i.e., how the intervention impacts the data market, how the data market impacts the intervention, and how their impacts interact. We study this interaction in two data-market settings to understand what information is necessary.
We find that without taking into account the incentive structure and economics of a data market, fairness interventions can induce greater losses to efficiency than are necessary to achieve fairness—even potentially inducing market collapse. Yet, we also find that these losses can be recovered or even amortized away by suitably designing the intervention with appropriate information or under favorable market conditions.
Overall, this thesis elucidates how data markets present both novel challenges and opportunities for machine-learning fairness. It demonstrates that efficiently intervening for machine-learning fairness can be more complicated in data markets—even infeasible! Excitingly, however, it also demonstrates that under favorable market conditions, fairness can be achieved at lower relative cost to efficiency than has previously been understood to be possible. We hope that these initial theoretical findings ultimately contribute to the development of efficient and practical fairness interventions suitable for real-world application
Considering Multidimensionality: A Solution for Noncognitive Assessment Application in Education
Full text linkAdvancements in education and the perceived necessary educational outcomes require advanced psychometric approaches. This is particularly true when we consider noncognitive assessments – assessments that aim to measure soft skills, otherwise untested in classrooms today. Noncognitive skills have gained traction in recent years due to the increased interest in 21st century skills and the realization that these skills impact success as much as technical skills and content knowledge.
Despite the growing interest, the appropriate psychometric methods have yet to be applied to such assessments. Because it is difficult to distill noncognitive assessments down to one particular skill, they are inherently multidimensional; however, literature regarding multidimensional applications of psychometric models is lacking.
This study aims to address these concerns, using Situational Judgment Tests (SJTs) to assess the feasibility of improving noncognitive assessment adoption across education. This study assesses multiple psychometric models, both uni- and multidimensional, to best understand the internal structure of SJTs as well as explore the interpretability of model results.
The results indicate that SJTs are, in fact, multidimensional and are best measured with polytomous Cognitive Diagnostic Models (CDMs). Implications of these results include future opportunities for noncognitive assessments to be utilized in classroom settings, increased feedback and reporting to students, teachers, and parents regarding strengths and weaknesses of students, and pedagogical improvements to noncognitive skill development
Demography, dynamics and data: building confidence for simulating changes in the world's forests
No full textVegetation demographic models (VDMs) are advanced tools for simulating forest responses to climate and land-use changes, and are essential for projecting carbon cycling and large-scale forest management strategies. Despite their increasing incorporation into Earth System Models, VDMs differ in their demographic assumptions, with no prior quantitative comparison of their performance.
We benchmarked nine VDMs against observational data from boreal, temperate and tropical sites, assessing their accuracy in predicting tree growth, carbon turnover, biomass stocks and size distributions. Models were simulated under consistent climate conditions with postdisturbance recovery monitored for at least 420 yr.
Postdisturbance carbon recovery trajectories showed significant variability while remaining within observational ranges. Initial regrowth rates varied substantially (0.03–0.60, 0.18–0.70 and 0.35–1.10 kgCm−2 yr−1 for boreal, temperate and tropical sites, respectively), influenced by each model's initial forest state. Models captured mature forest carbon content but showed compensating effects between overestimated growth and underestimated mortality rates.
This first multi-model benchmarking identifies growth and mortality rates as critical calibration targets and highlights the need to refine postdisturbance establishment conditions for model development. We outline specific benchmarking variables needed to improve predictions of forest responses to environmental change
Latinidad in Precision Medicine: The Boundaries and Extensions of Ethnic Identity in Biomedical Research
No full textBackground and research questions.
This dissertation analyzes the social processes and effects of deploying “Latinx” as a crucial population for advancing precision medicine research and practice. Precision medicine is an approach to healthcare that tailors treatment based on genetic understandings of disease and individuals. Sociologists have established that racialized concepts of human, and therefore genomic, differences persist as the operating logic in biomedicine, yet the relevance of race and ethnicity in genomics is contested, and there exists no standard definition of ethnic or racial categories within precision medicine. The questioned relevance of race and ethnicity in genomics is especially salient for the broad ethno-racial social category of Latinx, which straddles racial and ethnic categories.
Thus, focusing on Latinx populations for precision medicine presents a paradox: given that “Latino/a” or “Hispanic” is also institutionalized as the prototypical “ethnic” (read “nonracial”) social category, how can this group function as a meaningful biological, and specifically genetic, category? How might the use of Latinx as a biosocial grouping for the purposes of precision medicine have broader social effects, both on the self-understandings of Latinx people, and on others’ understandings of Latinx as a social and political category? Considering this and the breadth and heterogeneity of the ethno-racial social category of Latinx, this study takes a critical look at how precision medicine investigators conceptualize (and therefore construct) attributes of the Latinx population as “genetically meaningful” through recruitment practices and analyses.
Study methods.
This qualitative study incorporates 34 key informant interviews with precision medicine experts who work with Latinx communities and Latinx data, as well as 40 lay informant interviews with Latinx community members. This work is supplemented by a text analysis of 100 precision medicine publications and ethnographic observations of 23 precision medicine events that aided in crafting the interview guides and understanding the landscape of precision medicine.
Results.
Expert interviews reveal how researchers struggled to categorize the Latinx population through socio-political and biomedical understanding of “Latinx” as a large and ever-changing category, and how these interpretations led experts to adjust and toggle in their methodological or pragmatic research approaches. Expert informants reveal their definitions of “Latinx” through the subdivision of the category into subcomponents they believed carried the essence of Latinidad, which include, but are not limited to, social and cultural factors. Importantly, in characterizing “Latinx genetics,” experts described Latinidad as consisting of continental and racialized “genetic groupings” in the context of “Latinx admixture,” therefore racializing the ethnic Latinx category. Ultimately, researchers shared that their work was motivated by their desire to increase Latinx representation in research. Across precision medicine disciplines, experts also conveyed that socio-structural factors far outweigh genetic impacts on population health, and some researchers were skeptical that precision medicine advances will help Latinx population given the massive disparities in access to state-of-the-art medical technology and care.
Moreover, while sharing their unique composite understanding of what made them Latinx, lay participants described their identity as related to colonial history, and as having complicated associations with colorism. Further, lay informants overwhelmingly endorsed socio-structural determinants of health over genetic understandings of Latinx health. While themes of socio-politics come up as important for experts and lay informants alike, the major point of departure between lay informants and experts is that my lay informants described their genetics as stemming from and revealing their identity’s social histories while never tying genetics to their sense of belonging to the Latinx category
Learning with Auxiliary Information: A Unified Framework for Robust Clinical Prediction in Healthcare
No full textMachine learning offers unprecedented opportunities to advance predictive modeling in healthcare, driven by the vast increase in available clinical data. Despite the promise of increased performance, its applications pose many challenges due to the nature of healthcare datasets. Standard prediction models often fail in this context due to high dimensionality, data imbalance, irregular longitudinal collection, and data fragmentation across study cohorts. This dissertation addresses these challenges by introducing a novel paradigm, Learning with Auxiliary Information, which offers robust strategies to mitigate the limitations of standard predictive models. Auxiliary information is defined here as any information often discarded or underutilized in traditional machine-learning settings.
This work develops and applies novel machine-learning methodologies to diverse healthcare datasets to demonstrate this paradigm. For predicting preeclampsia in the Nulliparous Pregnancy Outcomes Study: Monitoring Mothers-to-be (nuMoM2b) cohort, ensemble methods were used to manage clinical complexity and address algorithmic fairness. To predict necrotizing enterocolitis (NEC) from neonatal stool microbiota, an attention-based Multiple Instance Learning (MIL) method was utilized on ambiguously labeled longitudinal data. A novel "growing bag" analysis was developed throughout an infant's early life to generate a dynamic, interpretable risk score for NEC. For predicting proximal junctional kyphosis (PJK) and preterm birth (PTB), where critical post-operative and delivery information is unavailable at inference, a new Learning Using Privileged information (LUPI) algorithm, XGBoost+, was created by integrating a distillation framework into gradient boosting. Each of the above applications demonstrates a unique way of handling auxiliary information. Finally, the overarching Learning with Auxiliary Information framework was instantiated by combining LUPI with Transfer Learning in a novel XGBoost+TL model. This demonstrated that knowledge could be successfully transferred from a large source dataset to improve prediction on a smaller, distinct clinical cohort.
The conclusions of this work confirm the viability of the proposed paradigm. The developed models consistently outperformed traditional machine learning approaches across all clinical problems. For preterm birth, the LUPI framework improved accuracy and revealed the starkly different predictability of indicated versus spontaneous preterm birth subtypes, providing key clinical insights. The successful application of XGBoost+TL confirmed that combining privileged information and transfer learning is a viable strategy for overcoming data fragmentation and scarcity. This dissertation concludes that moving beyond standard learning methods to a unified framework that strategically incorporates auxiliary data makes it possible to create significantly more powerful and reliable predictive tools to support high-stakes clinical decision-making