49755 research outputs found
Sort by
Dynamics of Future Soil Moisture Drought in Southwest North America: Linkages across Seasons in the Ocean–Atmosphere–Land System
Get PDFSouthwest North America is projected by models to aridify, defined as declining summer soil moisture, under the influence of rising greenhouse gases. Here, we investigate the driving mechanisms of aridification that connect the oceans, atmosphere, and land surface across seasons. The analysis is based on atmosphere model simulations forced by imposed sea surface temperatures (SSTs). For the historical period, these are the observed ones, and the model is run to 2041 using SSTs that account for realistic and plausible evolutions of Pacific Ocean and Atlantic Ocean interannual to decadal variability imposed on estimates of radiatively forced SST change. The results emphasize the importance of changes in precipitation throughout the year for declines in summer soil moisture. In the worst-case scenario, a cool tropical Pacific and warm North Atlantic lead to reduced cool season precipitation and soil moisture. Drier soils then persist into summer such that evapotranspiration reduces and soil moisture partially recovers. In the best-case scenario, the opposite states of the oceans lead to increased cool season precipitation but higher evapotranspiration prevents this from increasing summer soil moisture. Across the scenarios, atmospheric humidity is primarily controlled by soil moisture: drier soils lead to reduced evapotranspiration, lower air humidity, and higher vapor pressure deficit (VPD). Radiatively forced change reduces fall precipitation via anomalous transient eddy moisture flux divergence. Fall drying causes soils to enter winter dry such that, even in the best-case scenario of cool season precipitation increase, soil moisture remains dry. Radiative forcing reduces summer precipitation aided by reduced evapotranspiration from drier soils
Architectures Leveraging Switched Capacitor Circuits for Emerging Wireless Applications
No full textSwitched capacitor circuits have become a ubiquitous building block in various analog and mixed signal circuit architectures. Recent advancements in time interleaved switched capacitor circuits have demonstrated their capability of realizing a variety of frequency responses. Particularly, a polyphase multipath switched capacitor circuit can realize the response of a narrowband bandpass filter (commonly referred to as an -path filter) or a true time delay (referred to as a quasielectrostatic delay) based on ratio between the charging time constant of the capacitor and the modulation time period. In this dissertation, we explore these switched-capacitor circuit architectures to find applications in modern day wireless systems.
Full-duplex wireless is an emerging communication paradigm which enables simultaneous transmission and reception within the same frequency band. Enabling wideband full-duplex systems would require cancellation of the strong self-interference signal leaking from the transmitter to the receiver within the same node. This requires self-interference cancelers which emulate the self-interference channel across the desired bandwidth. -path based true time delay circuits present an attractive method to realize the self-interference canceler by utilizing multiple of these delay elements in parallel, creating a configurable FIR filter to adapt to the self-interference channel.We utilize this FIR filter to realize a time-domain RF canceler demonstrating wideband self-interference cancellation.
-path filters can be utilized to create compact, tunable, high-quality on-chip bandpass filters which offer an attractive alternative to bulky SAW filters in RF frontends. Typically, -path filters remain restricted to sub-6 GHz operation and low frequency selectivity. To circumvent this issue, we present high order -path filters operating up to 12 GHz frequencies by utilizing higher order intermediary baseband loads with overlapping clock signals to enable higher frequencies of operation.
Further, a transmission line periodically loaded with a switched capacitor can be utilized to realize time-interfaces in electromagnetic signals, which are the temporal analogue of spatial interfaces.We use CMOS switches to demonstrate an electromagnetic time interface operating inthe GHz-frequency range which can be utilized towards building time metameterials and photonic time crystals, opening a wide range of opportunities in the rising field of time-varying photonic media
Embodying Belonging: Gesture, Dress, and Diasporic Family Communication in Educational Spaces
No full textThis paper explores how gesture, dress, and other embodied practices serve as powerful forms of communication and cultural transmission among Black diasporic families in educational spaces. Drawing from linguistic anthropology, cultural psychology, Black feminist theory, and spirituality in education, the paper argues that schools often misread or pathologize embodied expressions rooted in cultural identity. Using a reflexive and interdisciplinary approach, the study illuminates how bodily literacy—such as tone, posture, and dress—functions as a method of meaning-making, resistance, and care. It calls for educators to develop an “embodied literacy” that recognizes the body as a valid site of knowledge. By re-centering embodied communication in the classroom, this work advocates for educational practices that affirm rather than alienate students and families whose cultural languages are written on the body
Toward Democratizing Interactive Data Interfaces
No full textInteractive data interfaces are critical in nearly every stage of data management—including data cleaning, wrangling, modeling, exploration, and communication. Yet creating new interfaces remains challenging—it requires careful specification of the analysis task, thoughtful interface design, and backend optimization to ensure responsiveness at scale. As data grows and shifts to cloud DBMSes, these challenges only intensify.
This dissertation envisions a more accessible and automated approach to building interactive data interfaces. It introduces DATA INTERFACE GRAMMAR, a compact and analyzable representation of interface analysis tasks. At the heart of DATA INTERFACE GRAMMAR is the idea of choice, which captures structural variations across queries while preserving their correspondence to interface design, thereby enabling new applications in automating both interface design and backend optimization.
To automate interface design, we develop PRECISION INTERFACES 2, the first system that automatically generates fully interactive data interfaces from query sequences, and NL2INTERFACE, a prototype that extends this capability to natural language queries. To automate backend optimization, we formulate the PHYSICAL VISUALIZATION DESIGN problem and develop JADE, a prototype system that automatically generates backend architectures satisfying both per-interaction latency requirements and resource constraints
Stimulated Raman Scattering Microscopy: Theory and Applications in Nano Imaging
No full textStimulated Raman scattering (SRS) microscopy is an emerging chemical imaging modality that has gained tremendous attention in biomedical and material science due to its fast label-free imaging capabilities. Despite significant achievements, the field of SRS imaging is largely driving empirically, leaving many fundamental questions unanswered or controversial. In particular, there has been disagreement regarding the enhancement factor and detectability when compared to spontaneous Raman scattering.
In this thesis, an alternative framework is presented to quantitatively understand SRS microscopy. Starting from a phenomenologically-defined stimulated Raman cross section (_SRS), the intrinsically molecular Raman response is revealed. Unlike the traditional spontaneous Raman cross section _Raman, _SRS turns out to be strong and even exceeding the electronic counterparts. _SRS is then connected with ?Raman through both a heuristic method and the full quantum electrodynamics derivation, which encompass both phenomena quantitatively in the same framework.
This new theory reveals a previously-unknown duality nature of Raman scattering, where both _Raman and _SRS can exhibit vastly different magnitudes for the same molecule, connected by the influence of vacuum zero-point fluctuations. This allows for the direct prediction of signal-to-noise-ratios (SNRs), vibrational population saturation, and photothermal effects. A mathematical model is built to discuss the fundamental detectability of both Raman techniques, which shows that SRS microscopy is almost always more sensitive than regular Raman microscopy. A diagrammatic approach reveals that SRS excels in high spatiotemporal regimes, explaining its advantage for microscopy applications.
Next, I will use the new theoretical frame to demonstrate the superiority of SRS microscopy in nano imaging, and apply the technique in nanoparticles. In particular, three examples will be presented, including solid lipid nanoparticles (Chapter 3), poly-lactic-glycolic-acid (PLGA) nanoparticles (Chapter 4), and polystyrene nanoparticles as an example of nanoplastics (Chapter 5). The imaging of these three types of nanoparticles points towards a strategy called generalized bio-orthogonal imaging, which fully utilizes the rich chemical information contained in Raman spectra for biomedical research
Essays in Asset Pricing and Private Equity
No full textThis dissertation includes three essays in asset pricing and private equity.
In the first chapter, Factor Model Selection Using the ICAPM, which is joint work with Paul Glasserman and Harry Mamaysky, we extend the factor model ICAPM consistency test of Maio and Santa-Clara 2012, using market data together with household-level consumption data. We find that more consistent factor models have less persistent alphas, and more stable betas and out-of-sample mean squared errors. We propose a novel statistical test for the sign of the consistency-stability relationship across many factor models and over time. Our methodology allows for the efficient identification of the historically most ICAPM-consistent factor models and factors from a large candidate pool. Our results suggest that historically consistent models are likely to be stable in the future.
In the second chapter, Venture Capital Risk: Theoretical Foundations, I provide theoretical foundations for VC-specific risk factors which capture domain-specific risk exposures of venture capital general partners (GPs). Particularly, I asks why GPs on aggregate make more intangible and more specialized investments during some times but not others in the context of a portfolio choice model, and what implications such aggregate investment behavior has for risk and returns. I argue that while making highly intangible or strongly sector-focused investments can prove profitable in good times, it simultaneously increases the potential for large losses during downturns. As the upside is tied to GPs’ expert knowledge, who can detect value in new technologies in ways which remain hard to replicate for outsiders, achieving it remains subject to asymmetric information leading to a limited willingness to share risk by less informed limited partners especially during downturns. Consequently, if such investments collectively amplify the cyclicality of the VC sector and endogenously increase risk during downturns, then rational GPs must be compensated on average for taking on such risk.
In the third chapter, Venture Capital Risk: Empirical Evidence, I empirically test the existence of VC-specific risk factors using deal-level VC returns. In order to achieve this, I introduce empirical proxies that capture the aggregate intangibility intensity of VC and the degree of technology sector specialization. Using these proxies and after accounting for public market exposure, I find evidence that GPs earn higher returns on average, when making more intangible or sector-focused investments, which is consistent with the theoretical arguments. Specifically, I find that startup investments associated with higher levels of intangibility and a strong sector focus tend to yield higher round-to-exit returns and are more likely to achieve successful exits through acquisitions or initial public offerings
Characterization and engineering of calcium-binding proteins for recovery and separation of rare earth elements
No full textRare earth elements (REEs), which include the 15 lanthanides plus scandium and yttrium, are essential components in permanent magnets, electronics, and green energy technologies. However, the separation of REEs remains a major industrial and environmental challenge due to their highly similar physicochemical properties. Conventional separation processes are chemically and energy intensive, generating significant waste. Proteins that natively bind calcium ions offer a promising bio-based platform for selective REE recognition due to the shared ionic radii and coordination geometries between REEs and calcium. This dissertation explores protein-based strategies to selectively bind and separate REEs using rational protein engineering, bioinformatics, and directed evolution.
In Chapter One, we investigated the REE-binding potential of the Block V of the RTX domain from the adenylate cyclase protein of Bordetella pertussis. This domain is intrinsically disordered and folds into a β-roll structure upon binding eight calcium ions. Using a FRET-based assay, we showed that the RTX domain binds trivalent lanthanide ions with significantly higher affinities than calcium, exhibiting apparent dissociation constants ranging from 20–75 µM. Notably, the domain demonstrates higher selectivity for heavy REEs over light REEs. Circular dichroism spectroscopy revealed that the domain undergoes pH-induced folding even in the absence of metal ions, suggesting protonation of key residues enables structure formation in acidic environments. Equilibrium ultrafiltration confirmed that the domain can coordinate up to four REE ions under extreme acidic conditions (pH < 1). Furthermore, we demonstrated practical application of the RTX domain by selectively recovering Nd and Dy from Fe and Co in a simulated NdFeB magnet leachate at pH 6.
In Chapter Two, we expanded our search for REE-binding proteins by conducting a bioinformatic analysis of calcium-binding peptides and domains. Seven unique candidates with distinct calcium-binding loop geometries were selected for experimental evaluation. This study revealed that the loop charge strongly correlates with REE affinity: highly charged, aspartic acid-rich loops demonstrated enhanced ion binding due to electrostatic repulsion and stabilization effects. Affinity trends across the lanthanide series favored ions with radii closest to calcium (~1 Å), consistent with evolutionary pressure for calcium selectivity. In solution, binding selectivity varied between proteins; however, immobilized proteins showed enhanced selectivity for intermediate REEs. One top-performing candidate, HEW5 from Nocardioides zeae, was immobilized in a 7 mL column and successfully achieved chelator-free, single-step separation of an equimolar La/Nd mixture. The separation yielded over 90% purity and 90% recovery. Moreover, immobilized HEW5 effectively removed non-REE ions from a simulated leachate stream and enabled selective separation of lanthanum from other REEs in a single chromatographic stage.
In Chapter Three, we addressed the need for rapid and efficient engineering of REE-selective proteins by developing a phage-assisted continuous evolution (PACE) system tailored for lanthanide binding. We constructed a selection circuit based on a lanthanide-mediated protein protein interaction, enabling high-throughput evolution of a calmodulin-derived peptide library. Within days, the system yielded a dominant evolved sequence with a restructured hydrogen bond network that enhanced second-shell ion coordination and overall protein packing. The evolved protein exhibited improved binding affinity and thermal stability, enabling high-purity, singlestage, chelator-free separations of individual REE ions. This work establishes a platform that can be extended to evolve binding domains for other critical metals, significantly expanding the scope of protein-based separation technologies.
In Chapter Four, we applied rational protein engineering to improve the selectivity and performance of the RTX domain. A truncated variant, termed RTX(2), was designed to reduce the heterogeneity of ion-binding sites. This engineered scaffold displayed enhanced REE binding affinity, greater structural stability, and pH-tunable elution characteristics. RTX(2) achieved over 200-fold selectivity between REE species during chromatographic separation, demonstrating improved resolution within the lanthanide series. This construct offers a robust, tunable platform for future protein-based separation technologies with direct applications in REE recycling and purification.
Together, these four studies demonstrate the power of combining natural protein scaffolds, bioinformatic screening, directed evolution, and rational design to address one of the most pressing materials challenges of the 21st century. This dissertation aimed to advance the development of protein-based REE separation platforms by revealing fundamental structure–function relationships and providing novel tools for scalable, sustainable metal recovery
The Hidden Forces of Rehab: Advancing Measurement Across the ICF Domains in Children with Unilateral Spastic Cerebral Palsy
No full textBackground.
Unilateral spastic cerebral palsy (USCP) impairs sensorimotor function primarily on one side of the body, limiting the ability to participate in daily activities. Improving rehabilitation requires scalable and inclusive measures that capture not only outcomes but also factors that shape the processes underlying change.
Purpose.
This dissertation addressed critical measurement challenges in pediatric rehabilitation by advancing tools and methods to quantify key mediators and outcome variables across the International Classification of Functioning, Disability and Health (ICF) domains for children with USCP. To advance knowledge in this area, four studies were conducted.
Results.
The first study (Chapter 2) described the development and validation of the Rehabilitation Observation Measure of Engagement (ROME), a video-based behavioral coding tool that quantifies both person-level and between-system level engagement during treatment. ROME demonstrated strong construct validity and reliability, with responsiveness to change across therapeutic activities.
Building on this tool, the second study (Chapter 3) used ROME to examine how child characteristics influence engagement and how engagement relates to functional hand outcomes. Findings indicated that age was significantly associated with engagement, and that engagement correlated with improvements in structured hand function tasks, highlighting its importance during capacity-based motor activities.
The third study (Chapter 4) addressed proprioception, which is commonly affected in children with USCP, by designing a multi-joint assessment protocol using a marker-based 3D motion analysis system. Children with USCP showed significantly reduced symmetry across all proprioceptive position sense (PPS) metrics compared to typically developing children (TDC). Results showed the importance of assessing PPS not only at the single-joint level but also across multiple joints, as orientation symmetry, which captures coordinated alignment across joints correlated significantly with functional outcomes.
The final study (Chapter 5) explored the use of a marker-less, deep learning-based algorithm as a scalable alternative to traditional motion capture for assessing motor behavior. This approach effectively quantified changes in gross motor movements and revealed intervention-specific trunk adaptations post-treatment. Fine motor applications still require further refinement.
Conclusion.
Collectively, these studies contribute to the body of tools for assessing engagement, proprioception, and motor function, enabling evaluation of intervention effects across ICF domains. This helps to advance both clinical assessment and research methodology in pediatric rehabilitation
New Advances in High-Dimensional Proximity Problems
No full textHigh-dimensional problems are increasingly prevalent across a wide range of scientific and engineering domains, including machine learning, computational biology, and data-driven decision systems. Solving these problems efficiently is of critical importance due to the overwhelmingly large volume of data that modern systems must process.
In many practical scenarios, it is not necessary to compute exact solutions; instead, proximity solutions are sufficiently close to the optimal one, which is acceptable and desirable. This relaxation opens the door to significant computational speed-ups.
In this thesis, we demonstrate four important high-dimensional proximity problems for which substantial acceleration can be achieved using advanced algorithmic techniques.
First, using the algebraic method, we design a faster average-case algorithm for the orthogonal vector problem and the closest pair problem. Both problems are high-dimensional problems, and they serve as cornerstones of fine-grained complexity, as one can refute SETH (strong exponential time hypothesis) by solving any one of them in subquadratic time.
Second, we develop a new method for approximate matrix multiplication, in which we can utilize any incomplete matrix multiplication tensors. We practice this new method and gain speed-ups on the correlation detection problem (i.e., the light bulb problem), a fundamental problem in learning theory.
Third, using geometric tools, we design the first high-dimensional Euclidean spanner with (1+ ) approximation and subquadratic size. Such a tool can embed high-dimensional vectors into a graph, enabling us to access all graph-based algorithms. We speed up the calculation of Earth-Mover distance via this tool.
Lastly, we design the world's fastest linear program solver. The key technical challenge is a dynamic matrix-vector multiplication maintenance problem. In our new algorithm, we combine many advanced techniques, such as robust central path, fast rectangular matrix multiplication, sketching algorithms, and multi-layer lazy updates on a dynamic data structure
Scaling Deductive Verification for Real-world Cloud Data Security
No full textThe use of sensitive private data in many applications from advertising to healthcare, often in the context of machine learning models, has raised concerns regarding the privacy of data in computing. These applications increasingly run on commodity cloud providers. For example, data and computation may be contained in virtual machines (VMs) running on shared hardware in the cloud, relying on a hypervisor to preserve VM isolation to protect applications and their data in VMs.
Although hypervisors and OSes are supposed to protect applications and their private data, their large codebases contain vulnerabilities that can risk data confidentiality and integrity. Vulnerable system software running at more privileged levels that can access application data is a significant security issue. Formal verification offers a potential solution to this problem by mathematically proving that system software can provide critical security guarantees. However, given the complexity of commodity systems software and modern hardware architecture, applying existing system verification techniques to fully verify a commodity system is infeasible.
This dissertation introduces a systematic approach to scale deductive verification for real-world systems and formally verify their high-level data security properties. It involves verifying that the software implementation satisfies a formal high-level specification of its behavior, then proving that the specification guarantees the desired security properties. It introduces novel verification techniques to modularize and automate the first part of this process -- proving that the software implementation satisfies the formal specification -- by security preserving layers and specification synthesis. It also introduces novel approaches to formally model and verify the data security properties in modern commodity systems.
To ensure the feasibility and solidity of the verification, it also demonstrates how to handle unmodified system code with advanced language features, how to model modern hardware features such as relaxed memory behaviors, cache and TLBs. We have applied these techniques to verify two real systems, SeKVM, a commodity multiprocessor hypervisor retrofitted from KVM, and Arm Confidential Compute Architecture (Arm CCA), a novel new architecture to protect the data security of virtual machines. All the proofs are implemented in Coq and are machine-checked