Rice University Theses and Dissertations

Permanent URI for this collection

https://hdl.handle.net/1911/13110

Rice University makes all graduate theses and dissertations (1916-present) available online at no cost to end users. Occasionally a thesis or dissertation may be be missing from the repository. If you are unable to find a specific dissertation, please let us know and we will attempt to make it available through the repository, provided that the author has not elected for it to be embargoed.

News

Visit the web site for Rice University's Office of Graduate and Postdoctoral Studies for more information about Rice graduate student requirements for theses and dissertations.

Browse

Now showing 1 - 20 of 14474

Embargo
Delivery of Large Gene Circuits In vivo Using an Engineered Baculovirus Vector for Multifactorial Control of Gene Expression
(2024-12-06) Brown, Lucas Bernard Clatanoff; Bashor, Caleb; Bao, Gang
Many of the viral vectors used for gene therapy are limited by the cargo size they can deliver into cells in tissue. As a result, most therapies being actively considered today tend to consist of monomodal expression of one or two genes. While this modality is undoubtedly effective for many applications, there remains advantages to being able to deliver more genetic cargo. A viral vector with an increased cargo capacity could allow room not only for more and larger therapeutic genes, but also regulatory elements that permit complex, multifactorial regulation of therapeutic gene expression. Here we use the insect-derived baculovirus capable of packaging and delivering >100 kb of transgene DNA as a vector for complex gene circuits that regulate and enhance in vivo gene therapy. Baculovirus has many advantages over other vectors: the ability to transduce a broad spectrum of mammalian cells, a large packaging capacity, no replication in mammalian cells, and a low toxicity in vivo. However, while baculovirus has been used as a gene therapy vector previously, its potential has been limited by its transient expression, as well as its susceptibility to inactivation by the complement system. We then implemented a hierarchical cloning scheme for the rapid generation and prototyping of baculovirus vectors containing up to 10 different expression units. We then address several shortcomings of the baculovirus by pseudo-typing the AcMNPV baculovirus with two proteins, the Vesticular stromatitis virus protein G and a fusion protein consisting of several complement regulatory domains. This engineered vector has increased transduction and persistence in mouse liver, muscle, and brain tissue. To our knowledge, this is the first time systemic delivery of baculovirus has been shown to be an effective delivery route. Using this engineered virus, we screened a library of 24 variations of a tamoxifen inducible circuit in order to select the architecture with the highest dynamic range, up to a 67-fold increase over uninduced. Finally, we demonstrate two orthogonal small molecule inducible systems (grazoprevir and tamoxifen) delivered by baculovirus in vivo, both as separate viruses and as one complete circuit. Our findings demonstrate the usefulness of complex regulation for the gene therapy field, as well as the utility of the baculovirus as a therapeutic vector.
Embargo
Motional Dynamics in Trapped Ions and Rydberg Atoms, and Hybrid Quantum Algorithm for Classical Optimization
(2024-12-06) Zhang, Zewen; Hazzard, Kaden RA
Quantum information science has emerged as one of the most promising fields in contemporary research, relying on both software and hardware innovations. This thesis looks for both algorithms with quantum features that provide advantages over classical algorithms, and better hardware platforms for experiments and quantum computing. The work spans theoretical studies in both algorithm and hardware design, including hybrid quantum-classical algorithms and the development of quantum information processing platforms. The algorithmic part has focused on the performance of a hybrid quantum algorithm - the Grover Quantum Approximate Optimization Algorithm (Grover QAOA) - designed for problems with multiple solutions. In practice, we find its potential for speedup in solution search and its ability to find all solutions. Furthermore, we propose a simplified protocol that reduces the classical complexity of optimizing the algorithm’s parameters, enhancing its practicality for future applications. Our implementation of Grover QAOA for multiple combinatorial optimization problems on trapped-ion quantum computers demonstrates that the algorithm can fulfill its fair-sampling advantage even on noisy devices. In the hardware part, we mainly explore how the motion of cold atoms can either be used to engineer interactions or lead to previously overlooked decoherence. The first hardware platform we discuss is trapped ions, where we focus on implementing individual addressing to natively simulate new types of many-body systems. Our proposal leverages the exceptional controllability of trapped ions to explore dynamical models such as topological pumping. The second hardware platform we study is Rydberg atom lattices, where we investigate the decoherence processes introduced by atomic motion during dynamics. Using the numerical tool of discrete truncated Wigner approximation, we simulate the coupled dynamics of electronic and motional degrees of freedom, demonstrating that atom motion induced by strong van der Waals interactions in Rydberg atoms can lead to significant decoherence in analog simulation experiments. We have also explored specialized topics involving other quantum hardware platforms. One area of study is the reduction of frequency crowding in superconducting circuit quantum chips. By properly designing the frequencies for each transmon qubit, we can improve the manufacturing yield of collision-free processors. Another area focuses on the SU(N) Fermi-Hubbard models on alkaline-earth-metal optical lattice platforms. We have obtained the phase diagram of unit-filling models with imbalanced spin flavors. This work aids future experiments in searching for potential ground states of the unit-filling SU(N) model.
Cosmological and astrophysical probes of axionlike particles
(2024-12-06) Hagimoto, Ray Mitchell; Long, Andrew J; Amin, Mustafa A
Axionlike particles (ALPs), pseudo Nambu-Goldstone bosons arising from the spontaneous breaking of global U(1) symmetries, appear in solutions to open issues in fundamental physics and are ubiquitous in string theory compactifications. Furthermore, ALPs have a rich phenomenology that provides numerous ways to search for evidence of their existence. This work explores two potential discovery channels for ALPs. The first considers the possibility that hyperlight ALPs, with masses less than 10^(-28) eV and a Chern-Simons coupling to electromagnetism, formed a cosmic string network in the early Universe that survives beyond recombination. In this scenario, cosmic microwave background (CMB) photons passing through string loops in the network experience a rotation in their plane of polarization, an effect known as CMB birefringence that may be within reach of future CMB probes. I use existing CMB birefringence power spectrum data to constrain axion string network parameters, then discuss non-Gaussian features of axion string-induced CMB birefringence maps, and finally explore how a neural network could estimate axion string network parameters from these maps. The second potential discovery channel examines how ALPs with lepton flavor-violating couplings and masses less than 1 MeV affect the cooling rates of neutron stars. Through these studies, I develop tools that would assist in identifying signatures of ALPs in cosmological and astrophysical observations.
Embargo
Reliable Medical LLM and Vision-Language RAG through Multi-Agent Orchestration and Single-Step Preference Alignment
(2024-12-06) Pahwa, Khushbu; Hu, Xia (Ben)
Medical RAG systems and long-context models like Med-PaLM face distinct yet interconnected challenges in processing complex medical information. While RAG systems struggle with hallucinations due to noisy retrievals and incomplete fact verification, long-context models, despite their ability to process extended inputs, suffer from attention dilution and context retention issues. Current BioNLP RAG systems have particularly overlooked the critical balance between retrieved context and parametric knowledge, often leading to hallucinations from over-reliance on retrieved information. Our HALO-MMedRAG framework addresses these challenges through a comprehensive multi-agent architecture. The system’s effectiveness stems from four innovative components: query-intent parser agent, multi-query generation agent, coarse retrieval agent, fine-grained hallucination aware retrieval agent with a perplexity and NLL based hybrid scoring for the chunks, generation agent, a light-weight fact-verification agent and an orchestrator agent that manages a CoT reasoning debate among 3 agents to provide the final hallucination free response, grounded in factuality. The notion of Retrieval Augmented Generation in the context of Multimodal Medical LLMs has not been given due consideration from the lens of hallucination mitigation. Further, the existing approaches have been limited in their coverage of the medical domains, often limited to X-Ray. Medical Multimodal LLMs, when utilized for Multi-Modal Retrieval Augmented Generation, face critical challenges in maintaining factual accuracy while integrating complex visual and textual information. Our innovative approach addresses these challenges through a unified Triple Preference Optimization framework with three-stage preference dataset curation, focusing on cross-modal alignment, retrieval balance, and a dual staged visual feedback agent. Unlike existing solutions, our method employs a single-step optimization process that simultaneously handles multiple aspects of alignment while maintaining computational efficiency. Through careful curation of preference datasets that capture different levels of alignment quality, combined with a visual feedback agent for precise visual grounding to provide visual prompting for the Vision Language Model to improve its response, our approach significantly reduces hallucinations and improves medical response accuracy. Extensive evaluation across diverse medical domains, including radiology, ophthalmology, pathology, magnetic resonance imaging and CT scan demonstrates superior performance compared to the existing multimodal medical RAG methods, making our solution titled Align-MedRAG-VL, both practical and reliable for real-world medical applications where hallucination mitigation is paramount.
Leveraging Graph Networks for Health and Wellbeing Prediction
(2024-12-05) Khalid, Maryam; Sano, Akane
Health and well-being prediction plays an essential role in mental healthcare and well-being-aware computing. The complex nature of well-being, resulting from its dependency on a person’s physiological health, mental state, and surroundings, makes its prediction a challenging task. In this work, we utilize mobile sensing data to predict self-reported well-being metrics. In addition to a person’s physiology, we incorporate the environment’s impact through weather and social network data. To this end, we leverage phone data to construct social networks and develop a machine learning architecture that aggregates information from multiple users within the graph network and integrates it with the temporal dynamics of data to predict well-being outcomes for all users. To address the dynamic nature of social networks, we introduce GEDD (Graph Extraction for Dynamic Distribution), an approach that automatically adapts to fluctuating network sizes. GEDD utilizes graph properties, including connectivity and components, to transform variable-sized graphs into a standardized format, ensuring no user data is discarded. The proposed architecture supports online learning, making it feasible to scale to large networks without adding ecological momentary assessments (EMAs) or additional data collection burdens, thus preserving user privacy. Through extensive evaluations, we show that social network incorporation improves prediction accuracy, although node influence, especially in users with high eigenvector centrality, can amplify noise. To address this, we propose a robust system that leverages attention and social contagion in well-being behaviors through graph networks and integrates it with physiological and phone data from ubiquitous mobile and wearable devices. This system is designed to predict well-being outcomes, such as sleep duration and other health metrics while mitigating the challenges posed by noisy and incomplete data. Finally, we further leverage the graph structure to reduce the user burden associated with collecting health and well-being metrics, which are often captured at a much lower resolution than sensing data through surveys and EMAs. To this end, we introduce a benchmark framework to evaluate existing state-of-the-art graph-based active learning (AL) strategies in dynamic sensing environments. Our framework assesses AL strategies in terms of adaptability to real-time, user-centric data by evaluating performance over time in a stream-based setting. We also introduce new metrics, including sampling entropy, coverage ratio, and time-gap analysis, to quantify user burden, sampling diversity, and generalization performance. These metrics provide a holistic view of the AL strategies’ effectiveness, helping to identify those that best balance predictive accuracy and user engagement. This comprehensive evaluation framework supports scalable and efficient health prediction systems, facilitating practical, large-scale deployment.
Embargo
A Radical Approach to Photochemical Transformations Using Earth-Abundant Metals
(2024-12-06) Bian, Kangjie; West, Julian
The continuing emergence of visible light-mediated photochemistry in modern organic syntheses has allowed facile access to powerful, unconventional reaction manifolds to synthesize diverse small molecules. Conventional photocatalysis/photoredox heavily relies on noble-metal based, coordinatively-saturated mononuclear photoactive complexes to perform a bimolecular outer-sphere single electron transfer (OSET) process for the generation of open-shell radical species. While powerful, the efficiency of this approach is limited by the bimolecular diffusion rate of reactants and photocatalysts and a redox-potential matching requirement for productive oxidative/reductive quenching. By contrast, direct coordination of a metal and substrate can offer a complementary reaction manifold by facilitating inner-sphere single electron transfer (ISET) to promote the homolytic cleavage of this metal-ligand bond and generate an open-shell radical, which can bypass the OSET redox-potential matching prerequisite. Light-induced ligand-to-metal charge transfer (LMCT) is such reaction manifold, allowing for selective single electron oxidation of the coordinated ‘ligand’. In general, this reaction scheme can convert anionic ligands to the corresponding radical forms which can function in various radical transformations including functionalization of unsaturated hydrocarbons and intermolecular C-H functionalization. Most importantly, a great synthetic advantage of this reaction manifold is that these processes are found in many early transition-metal (3d metals), which are significantly more earth-abundant than noble metals (e.g. Ir, Ru) used in traditional photoredox catalysts, presenting a low cost and sustainable alternative to noble metal photocatalysis. Apart from our exploration in LMCT catalysis, we also demonstrated radical ligand transfer (RLT) as an effective pathway to sequester transient alkyl radical species, introducing a powerful tool to utilize these reactive species for enhancing molecular complexity of feedstock chemicals. Herein, I will share my research of radical photochemical transformations enabled by earth-abundant metals and we hope our study of earth-abundant metal photocatalysis can inspire chemists to design sustainable pathways in pharmaceuticals and natural product syntheses. Radical difunctionalization is a powerful reaction scheme to incorporate useful functionalities onto unsaturated hydrocarbons, especially the prevalent unactivated alkene class. While atom transfer radical addition (ATRA) has been adopted in difunctionalization of unactivated alkenes to perform haloalkylation using the halide from the alkyl halide reagents, a more versatile reaction scheme that allows the incorporation of other functionalities by leveraging in situ generated transient alkyl radical intermediates is desirable. In chapter 1, we proposed bio-inspired radical ligand transfer (RLT) for taming transient alkyl radical species generated from radical addition to unactivated alkenes. Learning from the radical rebound process of the cytochrome P450 enzyme and non-heme iron-dependent oxygenases, we developed RLT catalysis to incorporate diverse functionalities to minimally functionalized alkenes. This efficient ligand transfer process outcompetes unproductive ATRA, indicating a powerful reaction manifold for functionalizing transient alkyl radical species. The RLT chemistry also inspires us to explore other alkene difunctionalization using earth-abundant metals. Vicinal diamine motifs are prevalent in bioactive molecules, pharmaceuticals, and molecular catalysts, underscoring their significance, and olefin diazidation has emerged as a promising strategy for synthesizing these motifs. Although synthetic precedents have utilized highly oxidative azidobenziodoxolone (ABX, Zhdankin reagent) or electrochemical methods to prepare this privileged motif, these protocols were often confined to limited substrate scope and procedural complexity. Motivated by our development of radical ligand transfer (RLT), we introduced the photochemical diazidation enabled by ligand-to-metal charge transfer (LMCT) and radical ligand transfer (RLT) in chapter 2. Leveraging the merger of these two reaction manifolds, we utilize a stable, earth-abundant, and inexpensive iron salt to function as both radical initiator (LMCT) and terminator (RLT) to synthesize valuable diazidated products. Mechanistic understanding of this cooperative LMCT/RLT also motivated us to develop a photocatalytic diazidation protocol and further expand this chemistry to photocatalytic dichlorination and regioselective fluorochlorination, suggesting the versatility of this tandem scheme. This cooperative scheme can also be applied to photocatalytic decarboxylative C-N bond formation, further demonstrating diverse nucleophilic reactants can be utilized in open-shell radical generation via LMCT, which can subsequently participate in cooperating reaction pathways. This cooperative system prompted us to explore sustainable photocatalysis, with goals of eliminating the usage of exogenous oxidants/reductant, driven by the cooperation of LMCT and other pathways. The introduction of fluoroalkyl groups to parent molecules is a powerful tool to modulate biological and physiological activities of these compounds through enhancement of lipophilicity, bioavailability and metabolic stability. One direct way to introduce these fluoroalkyl groups is through hydrofluoroalkylation of alkenes. Early studies have explored the utilization of expensive and/or oxidative fluoroalkylating reagents and precious metals, which significantly restrict the application of these strategies. In many respects, fluoroalkyl carboxylic acids are the most ideal fluoroalkylating source due to low cost and availability, with trifluoroacetic acid (TFA, $9/mol) as an example representing a desirable CF3 source for hydrotrifluoromethylation. However, the decarboxylation of TFA (and other fluorocarboxylic acids) is known to be extremely challenging due to its high oxidation potential, with previous approaches tentatively utilizing TFA confined to pre-installation of redox moieties to overcome this barrier, drastically decreasing the atom/step economy in these methods. In chapter 3, we show how leveraging the synthetic advantage of LMCT, which can override OSET redox-potential mismatching, allows us to develop a photocatalytic hydrofluoroalkylation protocol using fluoroalkyl carboxylic acids including TFA and other feedstock fluorocarboxylic acids enabled by cooperative LMCT and hydrogen atom transfer (HAT). Critical to the success is the cooperation of earth-abundant iron LMCT and redox-active thiol HAT, offering a mild and redox-neutral protocol to synthesize fluorine-containing molecules without the preactivation of feedstock fluoroalkyl acids. Following the development of photocatalytic hydrofluoroalkylation, we took inspiration from our development of photocatalytic diazidation and dichlorination, reasoning these (pseudo)halide X-type ligands could be equally applied in an analogous hydrofunctionalization reaction manifold. Exciting, we found this to be true and further expand cooperative LMCT/HAT to achieve a photocatalytic anti-Markovnikov hydrochlorination of unsaturated hydrocarbons. Enabled by selective oxidation of anionic chloride using weakly oxidizing iron to promote LMCT reactivities, previous challenging substrates are tolerated in our protocol, giving high anti-Markovnikov regioselectivity. Moreover, this hydrochlorination strategy can be applied to diverse alkynes, offering facile routes to preparing alkenyl chlorides in high regioselectivities and good stereoselectivities. Additionally, with simple adjustment of deuterated co-solvent, both deuterochlorination of alkenes and alkynes behave well in our redox-neutral system, providing another strategy of isotopologue syntheses. Lastly, this cooperative LMCT/HAT also inspires us to develop photocatalytic hydroazidation where we observe a critical ligand-acceleration effect. The facile photochemical generation of azidyl radical also allows us to explore LMCT in combination with halogen atom transfer (XAT) to develop regioselective haloazidation. These azidation protocols can address previous limitations in oxidative/corrosive reagent usage, high loading of metal sources, or limited substrate scope. Importantly, the cooperation of iron LMCT and thiol HAT has showcased a mild and general solution to hydrofunctionalization of unsaturated hydrocarbons. In this thesis, I have investigated the photochemical transformations enabled by earth-abundant metals, exploring the process of radical ligand transfer (RLT), ligand-to-metal charge transfer (LMCT), hydrogen atom transfer (HAT) and most importantly, the cooperation of these reaction manifolds which allows diverse transformations, establishing earth-abundant metal (photo)catalysis as a competitive synthetic manifold in accessing molecules of high value. These studies have been enabled by increasing mechanistic understanding of each reaction, fueling our continuous efforts in earth-abundant metal catalysis. We hope these studies enabled by earth-abundant metals communicate the importance of promoting sustainable (photo)catalysis in synthetic chemistry and serve as a powerful tool to synthetic chemists. We expect sustainable metal photocatalysis will keep enabling exciting chemistry!
Taming Data and Transformers for Audio Generation
(2024-12-05) Haji Ali, Moayed; Ordonez-Roman, Vicente
Generating ambient sounds is a challenging task due to data scarcity and often insufficient caption quality, making it difficult to employ large-scale generative models for the task. In this work, we tackle this problem by introducing two new models. First, we propose AutoCap, a high-quality and efficient automatic audio captioning model. By using a compact audio representation and leveraging audio metadata, AutoCap substantially enhances caption quality, reaching a CIDEr score of 83.2, marking a 3.2% improvement from the best available captioning model at four times faster inference speed. Second, we propose GenAu, a scalable transformer-based audio generation architecture that we scale up to 1.25B parameters. Using AutoCap to generate caption clips from existing audio datasets, we demonstrate the benefits of data scaling with synthetic captions as well as model size scaling. When compared to state-of-the-art audio generators trained at similar size and data scale, GenAu obtains significant improvements of 4.7% in FAD score, 22.7% in IS, and 13.5% in CLAP score, indicating significantly improved quality of generated audio compared to previous works. Moreover, we propose an efficient and scalable pipeline for collecting audio datasets, enabling us to compile 57M ambient audio clips, forming AutoReCap-XL, the largest available audio-text dataset, at 90 times the scale of existing ones. Our code, model checkpoints, and dataset will be made publicly available upon acceptance.
Embargo
Rad5 Replication Fork Rescue Mechanism Elucidation with a Structural Perspective
(2024-12-06) Molina Limon, Noel; Vlassakis, Julea; Gao, Yang
Stalled replication forks are frail structures with exposed single stranded (ss) DNA regions that are prone to digestion breaks. These are highly cytotoxic and energetically expensive to repair. Organisms have systems in place to protect and resolve stalled forks so replication can continue. Failure with said systems is closely associated with cell death and cancer. Fork reversal is a rescue mechanism for stalled replication forks that as the name implies, relies on re-zipping of the opened parental strand until a four-way double stranded intermediate is obtained. Helicase-like Transcription Factor (HLTF) in humans, and its low eukaryote ortholog Rad5, are prime examples of fork reversal enzymes. Interest in HLTF and its role in oncogenesis has been hampered by the complexity of expressing and manipulating this frail enzyme. Despite the hurdles, understanding the fork reversal mechanism is critical to identify how organisms discern between rescue events and disease progression. Exploiting the thermally resilient proteome of C. thermophilum, and functional homolog Rad5, allowed the study and elucidation of mechanistic details of enzyme-mediated fork reversal. Validation of Rad5 from a non-model organism exhibited the advantages of thermally stable enzymes and cements the future use of alternatives to overcome physical and logistical issues that plague recombinant enzyme production. Concerted biochemical and structural studies were performed to gain mechanistic information on Rad5 reversal of the replication fork. Our findings showed a regulatory effect by the enzyme’s distal n-terminal domain, that is linked to substrate specificity. Our work also identified the strand Rad5 translocating of DNA occurs on, which further solidifies our hypothesized mechanism and counters the enzyme’s historical description as an annealing helicase. Finally sample optimization and complex formation studies lay the groundwork for continuing structural tests.
How Gender and Race Influence the Relationship Between Pay Transparency and Negotiation
(2024-12-04) Argueta-Rivera, Jazmin; King, Eden
There is a persistent wage gap between women and men of color in the United States. Pay transparency (i.e., the degree to which an organization shares pay information) can expose pay imbalances between workers, make employees aware of injustices, and encourage pay negotiation. This is particularly important because people from marginalized backgrounds (White women and people of racial minority groups) tend to negotiate less than White men (Babcock & Laschever, 2003). I reason that pay transparency in job advertisements can be an opportunity for women and men of color to develop a sense of justice and trust toward the organization, ultimately emboldening their pay negotiations. In this paper, I explored differences in negotiation intentions stemming from pay transparency in job advertisements as explained by perceptions of distributive justice and organizational trust. In this experimental study, I did not find distributive pay transparency to be a precursor to negotiation, but I found evidence for the importance of trust in the propensity to negotiate for women job applicants.
Embargo
Long-Range, Large Aperture Thermal Imaging via Sparse Aperture Metalens Computational Imaging
(2024-12-03) Wang, Jing; Veeraraghavan, Ashok
Long-range imaging in the mid-wave infrared (MWIR) is critical for defense, industrial, and environmental applications, often requiring high-resolution imaging achievable only with large-aperture lenses (~100–1000 mm). Conventional glassbased refractive optics meet these equirements but result in bulky, costly systems. While metalenses offer a lightweight alternative, creating large apertures with current fabrication techniques poses significant challenges. To address this, we employed a computational imaging approach using sparse aperture metalenses. By arranging an array of small metalenses in a spatial configuration that maximizes spatial information and applying a computational reconstruction algorithm, our system achieves high-resolution, high-contrast images equivalent to those from a single large aperture. This scalable approach allows large-aperture realization by strategically arranging smaller sub-apertures. We validated this design with a prototype of 89 mm outer diameter and 356 mm focal length. Enhanced with a neural network-based reconstruction algorithm, our proposed sparse aperture system achieves near-diffraction-limited performance. Furthermore, simulations of a large recursive sparse aperture demonstrate improved MTF coverage and higher resolution imaging. This work represents progress toward practical, scalable, high-performance MWIR imaging systems.
Towards Fine-Grained Isolation Mechanisms for Intraprocess Isolation
(2024-12-05) Yang, Fangfei; Cox, Alan L.; Dautenhahn, Nathan
Memory safety has long been a significant challenge in computer software security. In this thesis, we propose a set of methods to mitigate memory safety issues. Our approach allows for isolation of different functions and modules within an application at the granularity of individual functions, thereby preventing the spread of memory safety issues between these modules. With our thread-safe security monitor, developers can specify untrusted code and data requiring extra protection, thereby restricting access to sensitive information in two key ways. The first, called a sandbox, isolates error-prone components, such as those used for computation, protocol state machines, and parsers. The second, called a safebox, protects sensitive data or security-critical elements, including privilege flags, access tokens, and ACLs. This model enhances data protection and supports the incremental isolation of critical parts at minimal cost. We introduce an innovative combination of memory safety with contextual re- sources, allowing the allocation of isolation contexts for temporarily created resources. For instance, this enables the isolation of communication contents between connec- tions from different users, with sharing permitted only through securely isolated mod- ules. A typical example is a chat server where each client has its own context for handling user connections and encryption keys, preventing attackers from accessing other users’ information. The received data is shared among receiving users through a shared memory within a safe module which include sufficiently small TCB code (containing only the minimal code required for memory copying). Finally, develop- ers can set additional system resource policies for these contexts, thus limiting their access to file systems, networks, and other resources. By utilizing alias mapping, we map the same physical memory to multiple vir- tual memory addresses, allowing different modules to share data structures without frequent copying. We embed this alias into the higher bits of the virtual address, which enables efficient address translation across domains with minimal overhead and facilitates the seamless use of shared memory across function calls. We simplify the process by allowing developers to express intentions rather than operations through code annotations, improving maintainability. This information can coexist with regular software code and be dynamically enabled or disabled through our tools, optimizing the use of limited hardware resources while balancing security and performance. Our system demonstrated 95% compatibility in LTP testing, indicating its ca- pability to support most applications developed for the Linux platform, including those that utilize signals and multithreading, without requiring additional porting. We conducted several micro-benchmarks for the implementation mechanisms our sys- tem relies on, better illustrating the system’s overhead sources and providing clearer guidance for users. We implemented module isolation in real applications like NGINX and Redis and created separate isolation contexts for user connections. These evaluations demon- strate that our system can be easily and progressively applied in practical software. Our overhead for individual module isolation ranged from 3% to 10%. When isola- tion was performed on both dimensions simultaneously, our overhead reached 10% to 40%
A kinetic Monte Carlo simulation of solid-electrolyte interphase formation and dendrite growth during electroplating
(2024-12-05) Cheung, Benji H; Tang, Ming
The formation of 3D structures such as dendrite, filament, and moss during electroplating is an obstacle to the development of a number of battery systems vital to a sustainable future, particularly lithium metal batteries. The morphological evolution of lithium metal electrode is strongly affected by the presence of passivating species formed by electrolyte decomposition, known as solid-electrolyte interphase (SEI). A 2D kinetic Monte Carlo (kMC) algorithm on a hexagonal grid was developed to account for the competing effects of deposition, diffusion, and surface passivation, providing an elementary understanding of electrodeposition systems with passivation. Growth from flat electrode and from hemispherical nucleus were both investigated. Morphological information and shape statistics were found to be strongly controlled by both SEI initiation time and current density, and a phase map was constructed over both parameters to demonstrate the distribution of results. Spherical deposits formed at high passivation time and high flux, filaments and whiskers at low flux and high passivation time, and dendrites and mosses at high flux and low passivation time. SEI formation is also observed to exacerbate nascent diffusion instabilities on pristine electrode. In the limit of no cross-SEI diffusion, we obtain a scaling relation of filament lengthscale as flux^(0.35) t_pass^(2.17). When cross-SEI diffusion was considered, a contrast is observed between low and high flux regimes: traditionally, thickness decreases with current, but at high fluxes after SEI cracking, we observe that growth from a single active tip can sustain larger thicknesses with larger flux, shedding light on an SEI-free ultrahigh flux regime. These findings provide foundational yet novel understanding of complex SEI phenomena, potentially streamlining the design of next-generation batteries.
Persistent tailoring of MSC activation through genetic priming
(2024-12-03) Beauregard, Michael Andrew; Diehl, Michael R
Mesenchymal stem/stromal cells (MSCs) are an attractive platform for cell therapy due to their safety profile and unique ability to secrete broad arrays of immunomodulatory and regenerative molecules. Yet, MSCs are well known to require preconditioning or priming to boost their therapeutic efficacy. Current priming methods offer limited control over MSC activation, yield transient effects, and often induce expression of pro-inflammatory effectors that can potentiate immunogenicity. Here, we describe a ‘genetic priming’ method that can both selectively and sustainably boost MSC potency via the controlled expression of the inflammatory-stimulus-responsive transcription factor IRF1 (interferon response factor 1). MSCs engineered to hyper-express IRF1 recapitulate many core responses that are accessed by biochemical priming using the proinflammatory cytokine interferon-γ (IFNγ). This includes the upregulation of anti-inflammatory effector molecules and the potentiation of MSC capacities to suppress T cell activation. However, we show that IRF1-mediated genetic priming is much more persistent than biochemical priming and can circumvent IFNγ-dependent expression of immunogenic MHC class II molecules. Together, the ability to sustainably activate and selectively tailor MSC priming responses creates the possibility of programming MSC activation more comprehensively for therapeutic applications.
Rapid Hydrogel Micropatterning and Cadherin Switching: Insights into Developmental Patterning and Mesendodermal Trajectories in Human Embryonic Stem Cells
(2024-12-06) Zhu, Ye; Warmflash, Aryeh
This thesis comprises two parts, each contributing novel insights into self-organized patterning and differentiation in human pluripotent stem cells (hPSCs). Self-organized patterning of mammalian pluripotent stem cells on micropatterned surfaces has been established as an in vitro platform for early developmental studies, complementary to in vivo animal models. The first project addressed the limitations of current micropatterning techniques, due to complex fabrication processes, such as micro-contact printing, preventing widespread usage in biological research. We developed a projection stereolithography-based micropatterning method that uses a digitally tunable photomask to rapidly print hydrogel with micro-features onto glass-bottomed-culture vessels. Combined with the laminin-521 (or LN521) extracellular matrix coating, this technology provides a surface suitable for hPSC attachment and growth with minimal non-specific cell adhesion. The study demonstrated the self-patterning results of hPSCs following gastrulation and ectodermal induction produced on our micropatterned surface are comparable with those obtained using commercially available micropatterned plates. This novel micropatterning approach enables customizable, rapid fabrication of micropatterned surfaces for cell study at a reduced cost, with potential application in developmental biology and regenerative medicine research. The second project explores cadherin switching during the epithelial-mesenchymal transition (EMT) in the differentiation trajectory of hPSCs through the primitive streaks (PS) and into mesodendermal subtypes. We measured EMT, and cadherin switching (E-cadherin downregulation and N-cadherin upregulation) during hPSC differentiation to PS and subsequently to distinct mesendodermal subtypes using established protocols and variants in signaling modulation of the key pathways, i.e., Activin, BMP, and Wnt. The findings reveal that while early signaling perturbations largely affected the extent of cadherin switching, the differentiation potential of PS cells was unimpacted. Specifically, definitive endoderm progenitors retained the ability to differentiate into both endodermal and mesodermal fates, while PS cells in mid to posterior regions exhibited restricted potential toward definitive endoderm. Additionally, E-Cadherin knockout did not alter cell fate outcomes in mesendodermal differentiation. Overall, the project revealed the decoupling of cadherin dynamics from cell fate decisions in mesendodermal differentiation through PS coordinates, with translational potential for cancer and age-related degenerative disease studies, where modulating EMT and cadherin switching could support innovative therapy development.
Embargo
Surveilling Im/mobilities Under the Digital Security State: An Ethnography of CBP One™ Across the Extended Mexico-US Borderlands
(2024-12-06) Flores, Lupe Alberto; Visweswaran, Kamala
The security state is expanding through digital border control and migration management. Based on twenty-eight months of ethnographic fieldwork in northern and central Mexico, I examine how the Mexico-US border becomes displaced across territorial and digital boundaries into various domains of social life along the migrant trail. Previous studies emphasize the political dimensions of surveillance within security states. By studying encounters between asylum-seeking migrants, humanitarian workers, US officials and software engineers acting under punitive asylum policies and digital innovation, my research interrogates how the sociotechnical development and implementation of CBP One™, a border-crossing smartphone app, generates bureaucratic, expressive and popular cultural forms that enforce and resist the transnational im/material bordering and surveillance practices of digital security states in the twenty-first century.
Embargo
Fast and Expressive Sketch Structured Transform for Efficient Inference
(2024-12-06) Saedi, Kimia; Shrivastava, Anshumali
Linear transformations using learned weights are fundamental components of deep learning models. Prior research has shown that dense weight matrices can often be compressed by decomposition, quantization, sparsification, or random parameter sharing without losing accuracy, suggesting the benefit of more efficient transformations. Among variants of weight matrices, structured ones have limitations in expressivity and quality-efficiency tradeoffs. Unstructured matrices are incompatible with modern hardware, leading to slower training and inference. To address these challenges, we propose Sketch Structured Transform (SS1), an expressive and hardware-efficient operator that reduces tensor multiplications and accelerates inference. SS1 leverages random parameter sharing in a block-structured manner, reducing computation while preserving the expressiveness of parameter sharing. We empirically show that SS1 achieves better quality-efficiency tradeoffs than competing variants. Our theoretical analysis also indicates that SS1 can be combined with quantization for further compression, and the experimental results confirm this. Additionally, pre-trained models can be projected using SS1 and finetuned for efficient deployment. Our experiments highlight various applications of the SS1, including (a) Training GPT2 and DLRM models from scratch for faster inference. (b) Finetuning projected BERT models for 1.31× faster inference while maintaining GLUE scores. (c) Proof of concept with Llama-3-8b, showing 1.11× faster wall clock inference using projected SS1 layers without finetuning.
Valentina V. for harp, immersive electronics, and lighting
(2024-12-05) Roy, Timothy Jordan; Lavenda, Richard
Valentina V. is an extended work for solo harp, immersive electroacoustic sound, and lighting, commissioned by Hope Cowan through a generous grant from the American Harp Society. The piece is inspired by my own musicological research into the medieval song “La harpe de melodie” and the provenance of the illuminated manuscript containing the song’s renowned pictographic musical score. Valentina V. is conceived as a tragic monodrama in which the solo harpist adopts the persona of 14th-century noblewoman and virtuoso harpist Valentina Visconti (1371–1408). Research suggests that the medieval song “La harpe de melodie” by Jacob de Senleches — famous for its illuminated pictographic score — was likely composed as a vehicle to showcase Valentina’s prodigious musical talents. Married to the brother of the King of France, Valentina was eventually forced into exile after others at the royal court accused her of witchcraft. My piece presents an imagined scene near the end of Valentina’s life in which she is confined to her chamber with only her precious harp to confide in. Cast in four movements played without pause, Valentina V. unfolds fantasia-like as its protagonist processes her grief through playing her beloved harp. Compositional materials are partly derived from “La harpe de melodie,” which is performed in full as the work’s penultimate movement. At other points in the piece, the song emerges in a fragmented, distorted, or embellished form, representing Valentina’s reminiscences as they are filtered through her fractured psyche.
Mechanistic Understanding of ML/AI Systems Through Interdisciplinary Scientific Applications
(2024-12-03) Ju, Yilong; Patel, Ankit B.
Modern artificial intelligence (AI) systems have achieved remarkable success across various scientific domains. However, fundamental questions remain about how these systems learn, make decisions, and generalize across different applications. This dissertation addresses these questions by systematically analyzing and improving AI systems through applications in physics, chemistry, and healthcare, demonstrating how mechanistic understanding can enhance practical performance. First, we develop a unifying framework for understanding convolutional neural networks (CNNs) in quantum physics applications. We show how CNNs efficiently approximate quantum wavefunctions in exponentially large Hilbert spaces using only linearly many parameters by connecting them to maximum entropy models and correlator product states. This analysis reveals how CNNs leverage quantum system symmetries and entanglement properties, leading to a new training algorithm that significantly reduces convergence time or number of parameters while maintaining accuracy. This work establishes a bridge between physics and machine learning, providing a template for analyzing other neural architectures and suggesting when they might succeed or fail in solving certain physics problems. Second, we develop a series of machine learning approaches for chemical spectroscopy analysis. The Characteristic Peak Extraction algorithm improves accuracy for identifying chemical components in complex mixtures, while our Characteristic Peak Similarity metric enables accurate matching between different types of spectroscopic measurements. These tools are being actively tested to detect harmful chemicals in environmental samples and human organs, including polycyclic aromatic hydrocarbons in soil and placenta samples. This work creates more accessible and efficient tools for environmental monitoring, addressing a longstanding challenge in the field of analytical chemistry where traditional chemical spectroscopy methods require extensive laboratory facilities, expert knowledge, and time-consuming analysis procedures. Finally, we advance medical diagnostics by creating interpretable deep learning models for ECG analysis that achieves high accuracy in detecting junctional ectopic tachycardia. Through explainable AI techniques, we systematically analyze how these networks make decisions by identifying key ECG features that align with clinical expertise, categorizing error patterns, and conducting root cause analysis of misclassifications. This mechanistic understanding not only validates the model's reasoning against clinical expertise but also provides insights for model improvement and clinical deployment. Beyond the immediate clinical impact, this contribution provides a framework for developing trustworthy AI systems in healthcare, where understanding decision-making processes is crucial for clinical adoption. Together, these contributions advance our understanding of AI systems while demonstrating their practical impact across multiple scientific disciplines. The frameworks and methodologies developed in this thesis provide a foundation for building more interpretable, efficient, and reliable AI systems.
Underwater Electric Arc Synthesis of Ammonia and Machine Learning Guidance for Synthesis of Antimicrobial Aminocyanines
(2024-12-06) Lathem, Alex Ean; Tour, James M
Increasing demand for ammonia is expected in future years due to its potential as an electrochemical fuel and continued use in growing food for billions of people. Meanwhile, there is a growing need for novel antibiotics in the face of antimicrobial resistance worldwide. In this thesis, novel synthesis methods to address both challenges are explored. First, a novel method of ammonia synthesis is demonstrated using a nitrogen stream running through an underwater electric arc. Variation in ammonia yield is shown for a wide array of parameters, including electrode material and geometric configuration. Yield and energy efficiency are compared to other prominent bench-scale ammonia synthesis techniques in the literature. Second, machine learning analysis is conducted on a dataset of cyanine-derived molecules and their inhibition of bacterial growth. The most performant model is blind-tested against additional data, and then promising candidate molecules are offered for future synthesis and testing.
Implementation and Use of the Neuromusculoskeletal Modeling Pipeline
(2024-12-02) Hammond, Claire Vivian; Fregly, Benjamin J
Neuromusculoskeletal injuries including osteoarthritis, stroke, spinal cord injury, and traumatic brain injury affect roughly 19% of the U.S. adult population. Standardized interventions have produced suboptimal functional outcomes due to the unique treatment needs of each patient. Strides have been made to utilize computational models to develop personalized treatments, but researchers and clinicians have yet to cross the “valley of death” between fundamental research and clinical usefulness. This article introduces the Neuromusculoskeletal Modeling (NMSM) Pipeline, two MATLAB-based toolsets that add Model Personalization and Treatment Optimization functionality to OpenSim. The two toolsets facilitate computational design of individualized treatments for neuromusculoskeletal impairments through the use of personalized neuromusculoskeletal models and predictive simulation. The Model Personalization toolset contains four tools for personalizing 1) joint structure models, 2) muscle-tendon models, 3) neural control models, and 4) foot-ground contact models. The Treatment Optimization toolset contains three tools for predicting and optimizing a patient’s functional outcome for different treatment options using a patient’s personalized neuromusculoskeletal model with direct collocation optimal control methods. Two NMSM Pipeline use cases are presented. The first example is an individual post-stroke with impaired walking function, where the goal is to predict how the subject’s neural control could be changed to improve walking speed without increasing metabolic cost. First the Model Personalization toolset was used to develop a personalized neuromusculoskeletal model of the subject starting from a generic OpenSim full-body model and experimental walking data (video motion capture, ground reaction, and electromyography) collected from the subject at his self-selected speed. Next the Treatment Optimization toolset was used with the personalized model to predict how the subject could recruit existing muscle synergies more effectively to reduce muscle activation disparities between the paretic and non-paretic legs. The software predicted that the subject could increase his walking speed by 60% without increasing his metabolic cost per unit time by modifying the recruitment of his existing muscle synergies. This hypothetical treatment demonstrates how NMSM Pipeline tools could allow researchers working collaboratively with clinicians to develop personalized neuromusculoskeletal models of individual patients and to perform predictive simulations for the purpose of designing personalized treatments that maximize a patient’s post-treatment functional outcome. The second example is a novel personalized closed-chain kinematic shoulder model creation process utilizing the first Model Personalization tool, Joint Model Personalization. Commonly used kinematic shoulder models typically use regression-based kinematics and open-chain constructions, these models can produce low accuracy and anatomically impossible kinematics for many motions and subjects. After creating synthetic marker data and a model compatible with the NMSM Pipeline, joint parameters were automatically optimized to minimize the error between modeled kinematics and experimental kinematics of eight motions. The software produced a series of models with average marker distance errors below 1 millimeter across all motions for the best 5 degree of freedom model. This novel personalized closed-chain kinematic shoulder illustrates the ability of the NMSM Pipeline to influence the field of neuromusculoskeletal modeling

Browse

Recent Submissions