80264 research outputs found
Sort by
DDA-based Inverse Design of Nanophotonic Metasurfaces: a Benchmarking of Optimization Techniques
Get PDFThis work is an in-depth assessment of the optimization techniques used for metasurface inverse design in the field of nanophotonics. Each technique is detailed and implemented, comparing and contrasting the results produced by the numerical methods. A unique DDA solver is used to speed up the computation of the electromagnetic fields, which are used to determine the quality of the proposed designs. Fabrication limitations are also taken into consideration, and various filters were used to introduce a minimum length-scale to the designs. Finally, multi-band optimization was implemented in order to prevent single-wavelength peaks, which are unrealizable in practice
Generative Language Models for Program Synthesis and Evaluation
Get PDFRecent advances in Large Language Models (LLMs), such as GPT and Claude,
have significantly advanced the field of program synthesis. To evaluate the performance
of these models, traditional benchmarks like APPS, MBPP, and HumanEval
reveal limitations due to potential data leakage and their inability to mirror the
complexity of real-world programming. These benchmarks typically feature concise,
stand-alone code samples that fail to assess the nuanced capabilities required for
comprehensive coding tasks adequately. To address these limitations, this dissertation
introduces a novel, private benchmark dataset - SimCoPilot, specifically crafted
to simulate the ability of an AI such as a large language model (LLM) to perform as a
“copilot”-style, interactive coding assistant. In SimCoPilot, an AI is asked to provide
small amounts of code within an existing project, ranging in size from hundreds to
thousands of lines. The benchmark tests an AI’s ability to write code in both completion
(providing code to finish a method or a block) and infill scenarios (providing
code to fill a blank in a method), covering various domains such as classic algorithms,
databases, computer vision, and neural networks.
Despite their varied architectures, most LLMs typically treat source code as mere
string objects and require large-scale models and extensive training datasets. Unlike
natural language, however, source code is a formal language imbued with rich syntactical and semantic structures. Addressing this disparity, this dissertation explored
an innovative approach that explicitly extracts and integrates these syntactic and
semantic elements into an encoder-decoder transformer model.
Our detailed evaluation analyzes how LLMs manage different code dependencies
and logic complexities, providing insights into their operational effectiveness in
realistic programming environments. This examination provides profound insights
into the capabilities of modern Language Models in navigating realistic programming
challenges, thereby making a significant contribution to the understanding of their
practical applicability in the software development environment
Understanding and Leveraging the Temperature-Dependent Curing of Silicone Elastomers
No full textSilicone elastomers offer a wide range of mechanical properties, and their inherent compliance renders them suitable for use in applications including medical devices, shock absorbers, water-repellant surfaces, and cookware. Moreover, in the past decade, silicone elastomers have facilitated significant progress in the field of soft robotics. However, knowledge of the curing duration at a given temperature for thermally polymerizable elastomers often relies on empirical trends, and furthermore, the curing parameters—typically determined through trial and error—are limited to specific geometries and elastomers. Additionally, over-curing elastomeric parts at elevated temperatures consumes excess energy and contributes to device failure due to subsequent weak adhesion between components. The lack of understanding of the curing behavior limits the accessible design space of elastomers. Building on a framework introduced in my prior research quantifying the inactivation reaction of viruses, in this thesis, I present a modeling framework based on thermo-rheological experiments and the Arrhenius equation to provide a new understanding of the temperature-dependent curing of platinum catalyzed elastomers. The experimental results reveal that the curing behavior exhibits self-similarity upon normalizing with the gelation time, and the reaction is characterized by a dimensionless reaction coordinate that represents the extent of curing. Next, I leverage this understanding of the curing kinetics to study the adhesion between elastomer layers only as a function of the extent of curing, which accounts for both duration and temperature, and demonstrate the utility of the reaction coordinate to pinpoint failure regimes. Adhesion between elastomeric components represents a longstanding problem in the field of soft robotics and soft lithography. New insight into improving adhesion will enable new fabrication methodologies. Finally, I investigate the effects of curing silicone elastomers at temperatures beyond room temperature on the mechanical properties. The corresponding experimental results highlight the feasibility of using temperature to control the speed of curing while maintaining the desired mechanical behavior. Overall, my thesis aims to understand and leverage the curing behavior of elastomers as a function of time and temperature, informed by reaction kinetics, to broaden elastomer processing beyond traditional casting, and expand the accessible design space for the manufacturing of elastomeric devices
MODELING AND OPTIMIZING POROUS ELECTRODES WITH PHASE TRANSITIONS AND INTERCALATION STRESS
No full textThis thesis investigates the optimization of lithium-ion battery electrode parameters and the utilization of advanced simulation techniques to enhance battery performance. Layered lithium nickel-manganese-cobalt oxide (NMC) and lithium iron phosphate (LFP) are the two most widely used cathode materials for lithium-ion batteries. Battery performance is controlled not only by the intrinsic properties of the active materials (e.g. NMC, LFP) but also the porous electrode structure at the battery cell level. This thesis deals with cell-level battery modeling and is structured in three comprehensive chapters. The first chapter discusses the application of high-efficiency computational tools for simulating lithium-ion batteries with NMC cathodes. It emphasizes the adoption of the Pseudo-Two-Dimensional (P2D) model and computational approaches to expedite parameter optimization, which is crucial for improving battery performance. The second chapter critically evaluates four modeling techniques for LFP cathode batteries, with a focus on their capability to accurately fit experimental data regarding the lithium diffusivity coefficient. The efficacy and precision of the selected models are thoroughly assessed. The final chapter explores the incorporation of phase-transition-induced stress in the model for LFP, analyzing how various model parameters influence the battery’s discharge curve and the reaction distribution. It includes a discussion on the current gaps in comprehensive analyses of phase transition materials and highlights the need for better computational models to simulate the stress effect during battery charge / discharge
How Gender and Race Influence the Relationship Between Pay Transparency and Negotiation
No full textThere 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
Deploying Usability: Ensuring Trust in Electronic Voting for Military Absentee Voters
No full textU.S. uniformed service members deployed overseas face unique challenges in exercising their right to vote, often showing lower voter confidence due to difficulties in updating registration, requesting absentee ballots, and meeting tight voting deadlines. The current research evaluated the usability and trustworthiness of “CACvote,” an absentee voting system designed to enhance voting access and security for military personnel. CACvote allows voters to instantly request ballots and verify the legitimacy of their mail-in votes using secure authentication. Through a series of three experiments, the research sought to determine whether military voters find CACvote intuitive and trustworthy.
First, a baseline experiment was conducted to establish usability and trust metrics with eligible military voters for a traditional electronic voting system (“Baseline”), simulating the core functions of electronic ballot-marking. Building on this model and insights from the experiment, a second formative experiment explored how additional features of the user interface—including support for authentication, ballot verification, and secure mailing processes—impact usability and trust. These findings informed subsequent system iterations. Finally, the third experiment evaluated the refined voting system’s overall usability and trustworthiness with military voters.
CACvote demonstrated increased vote-casting rates and reduced assistance requests compared to the Baseline system, while maintaining similarly high levels of user satisfaction and trustworthiness. No significant differences were found in perceived workload or satisfaction between the two systems, suggesting that CACvote’s novel features integrate effectively with traditional electronic voting methods. This research also explored the relationship between trust and voting system usability, highlighting the distinct roles of past voting experience and perceived workload in shaping the voter’s trust in the voting system. These findings contribute to the exploration of how voting system usability influences voter confidence, highlighting the role of trust as a key indicator of usable security
Adapting learning and search algorithms to handle protein structural data with the goal of aiding drug discovery
No full textExperimental methods for protein structure determination (e.g., x-ray crystallography, NMR, cryoEM) require access to expensive equipment and are not scalable. Computational methods assist protein structure prediction and analysis on a far larger scale. Recent deep learning advances, the most notable being DeepMind’s AlphaFold2.0 release in 2021, have provided a wealth of structural data for further analysis and open new opportunities for algorithmic development. In my work, I address three different tasks that make use of the available protein structure data: (1) system-specific binding-affinity prediction (in the context of the immune-related peptide-HLA system); (2) generation of representative ensembles from generic protein structure datasets; (3) protein-ligand ensemble docking. To this end, I examine and adapt a range of algorithms including random forest regression models, unsupervised learning methods and stochastic global optimization techniques. I validate the resulting pipelines on available experimental data and apply them to different macromolecular contexts such as the immune-related formation of the peptide-HLA complex; flexibility of the signal transducer PI3K lipid kinase; CDK2 protein kinase and estrogen receptor α. Developed pipelines are open source and freely available and can help guide the search for novel therapeutics
Underwater Electric Arc Synthesis of Ammonia and Machine Learning Guidance for Synthesis of Antimicrobial Aminocyanines
No full textIncreasing 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
Mechanistic Understanding of ML/AI Systems Through Interdisciplinary Scientific Applications
No full textModern 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
Dataset
No full textThe manuscript was published in Nature Communications in 2025. The preprint is available at https://arxiv.org/abs/2410.21171.
The dataset contains the COMSOL simulation data for the figures in the main manuscript, including the normalized power spectra for different polarizations, and the |E_-| profile and the ellipticity profile. This includes the simulations for the linear cavity and designs I-IV discussed in the manuscript.A work proposed a scheme to realize terahertz chiral photonic-crystal cavities with broken time-reversal symmetry and investigated the Dirac gap opening in graphene when coupled to the chiral cavities