48440 research outputs found
Sort by
Ultrafast dynamics of charge carriers in 2D chalcogenides and the impact of zero-valent metal intercalation
No full textTwo-dimensional (2D) materials, characterized by strong in-plane covalent bonding and weak out-of-plane van der Waals interactions, exhibit unique electronic and optical properties. This work focuses on GeS, a group-IV monochalcogenide, and SnS2, a transition metal dichalcogenide, which have garnered interest due to their high carrier mobility, air stability, earth abundance, in-plane anisotropy, and strong spin–orbit coupling. The bandgaps in the visible-to-near-infrared range make them promising candidates for solar energy conversion applications, and observation of the shift current, a type of a bulk photovoltaic effect (BPVE) holds promise for photovoltaic devices that can surpass the Shockley-Queisser efficiency limit. Additionally, their properties make them suitable for energy storage, chemical sensing, photodetection, and ultrafast optoelectronics. This study explores the electronic and optical properties of GeS and SnS2 and investigates their tunability through intercalation. Intercalation introduces external species into the van der Waals gaps, modifying material properties without disrupting the host lattice. The intercalation of zero-valent Cu atoms and organic molecules into GeS, as well as the intercalation of SnS2 with Ni, Bi, Rh, Cu, and Fe, is examined. In this work, two primary experimental techniques were utilized: (1) time-resolved THz spectroscopy (TRTS) probes microscopic transient frequency-resolved photoconductivity following optical excitation and (2) THz emission spectroscopy (TES) investigates nonlinear optical effects and ultrafast photocurrents. Using TRTS, carrier dynamics in GeS nanoribbons were studied under 400 nm photoexcitation, which primarily excites surface layers, and 800 nm, which penetrates the bulk and injects carriers into the lowest conduction band valleys. Surface and bulk-excited carriers exhibit distinct lifetimes and mobilities, with surface carriers primarily influenced by surface defects. Additionally, intercalation of Cu0 was found to reduce the lifetimes and enhance the mobility of bulk photoexcited carriers in GeS. In SnS2, TES study revealed that inversion symmetry breaking induced by stacking faults enables 2nd order nonlinear effects, such as optical rectification and shift current, leading to the emission of THz pulses and establishing SnS2 as a potential emitter material for THz photonics. Intercalation of SnS2 with zero-valent metal atoms, including Fe, Ni, Cu, Bi, and Rh, enhanced optical absorption and altered photoexcited carrier dynamics, highlighting intercalation as an effective strategy for tailoring the electronic and optical properties of 2D semiconductors for applications in ultrafast optoelectronics, solar energy conversion, and photonics technologies
Viability of Timber-Concrete Composites for Outdoor Pedestrian Bridges
No full textMass timber is a family of engineered wood products that has risen in popularity in recent years due to its high strength performance, warm aesthetics, and potential as a renewable, low-carbon building material. One application for the use of mass timber is in timber-concrete composites (TCC). There are numerous arguments for adapting TCC technology into outdoor pedestrian bridge applications including high strength and dynamic performance, durability, and environmental impacts. Despite the purported benefits of TCC in outdoor pedestrian bridges, research in the USA has not developed enough to permit codification. The purpose of this research was to address apparent gaps in TCC literature and propose a model for the design of outdoor TCC pedestrian bridges in the USA. Key areas of investigation included the development of a framework model for assessing the strength, serviceability, dynamic performance, and durability of a TCC bridge through spreadsheet analyses, finite element models, and diffusion models. Finally, an embodied carbon and material cost estimate were developed, and a comparison was made with an equivalent steel-concrete composite pedestrian bridge. It was found that timber-concrete composites are a promising solution for outdoor pedestrian bridges with a high degree of structural efficiency, durability, and environmental-economic benefit. Frameworks were developed to identify best practices related to structural analysis, diffusion modeling, and durability detailing. Future research to expand on this study could include full scale mock-ups to evaluate dynamic performance, investigations with different shear connectors, and evaluations of crash tested railings to enhance material efficiency
Machine Learning-Based Estimation of EMG Baseline Noise Standard Deviation Without Rest Data
No full textAccurate estimation of baseline noise standard deviation in surface electromyography (EMG) is essential for EMG amplitude estimation, particularly during low-level contractions where signal-to-noise ratios are low. Traditionally, rest-state EMG recordings are used to estimate background noise standard deviation, but such data are often unavailable. This study introduces a machine learning approach that estimates resting-state EMG noise standard deviation directly from active contractions, removing the need for explicit rest trials. A hybrid convolutional-recurrent neural network was trained on physiologically realistic simulated EMG and fine-tuned on experimental recordings from three different EMG acquisition systems from a combined total of 62 subjects. Model performance was evaluated using EMG from constant-force and force-varying contractions about the elbow. Direct rest based noise standard deviation measurements remained the most precise (absolute difference inter-trial median and inter quartile range (IQR) of 0.06%, 0.18% maximum voluntary EMG-MVE), whereas machine learning provided meaningful improvements when rest data were unavailable (median and IQR absolute difference from rest-based noise standard deviation of 1.4% MVE, 1.88% MVE). In a separate evaluation of baseline noise reduction, omission of noise correction performed worst, machine learning performed better (noise reduction of 45% compared to no noise correction), and noise calibration from a rest contraction performed best (noise reduction of 75% compared to no noise correction); with all differences being statistically significant. These results demonstrate that machine learning offers a viable alternative for EMG noise correction in settings where rest trials cannot be collected, enabling more reliable EMG interpretation
Coordination in Multi-Agent LLM Systems: The Role of a Question-Asking Agent in Guiding Collaborative Consensus
No full textCoordination, debate, and reflection have shown promising improvements in multi-agent Large Language Model (LLM) task performance. Inspired by the role of questioning in human group reasoning, this research introduces a novel component to multi-agent LLM systems: a Question-Asking Agent (QAA) that guides collaboration through targeted, uncertainty-reducing questions. The QAA selects questions based on Expected Information Gain (EIG), a metric used to quantify the value of information a question may provide. To evaluate the impact of the QAA, a multi-agent LLM system was implemented and tested on the chess game state tracking task, a benchmark problem that challenges LLMs to maintain consistent reasoning across a sequential input. The system included generic agents collaborating through dialogue and a QAA generating questions using template-based formulations with calculable EIG. Experiments were conducted across 15 configurations varying the number of agents (1-5) and QAA strategy (none, random, EIG-driven). Results show that the QAA with EIG consistently improved system accuracy compared to both the baseline and the random-question QAA. Additionally, increasing the number of agents showed improvements across all QAA strategies. This study demonstrates that EIG-guided questioning can significantly improve reasoning performance in multi-agent LLM systems. These findings open new directions for enhancing coordination, interpretability, and performance in multi-agent LLM settings across a range of structured reasoning tasks beyond chess
Design and Development of Intelligent, Low-Power, Wireless Wearable Sensors for Biopotential Measurement
No full textIn the first part of this thesis, a Bluetooth Low Energy (BLE) based wireless biosensor network was investigated to determine its feasibility in a wearable EMG-driven prosthesis controller. In this investigation, the BLE connection parameters of connection interval and event length as well as the distance between the central and peripheral nodes were varied to study the biosensor network’s performance in terms of latency, power consumption and number of supported connections. For prosthesis/orthosis control, the system latency should be well below 100 ms to avoid noticeable lag in the device’s response. It was found that the BLE test system was able to achieve an average latency of just over one connection interval (10–20 ms in this research) with a maximum of just over two. The system will be battery powered and is intended for 12 hours or more of use, so the average currents should be no more than a few milliamps. The measurements found that for all cases considered, the average current consumption was under 3 mA. Finally, some prosthesis/orthosis controllers may require multiple sensor nodes to enable control from different locations. This investigation found minimizing the event length will allow for more sensor node connections at a given connection interval setting and that increasing connection interval also increases the number of possible connections. The results of this investigation laid the foundation for a take-home case study in which the wireless system was tested as part of a prosthesis controller. In the second part of this thesis a machine learning based framework was proposed for ECG interpretation and lead reconstruction in a resource constrained environment. A lightweight CNN-based ECG classification model was developed and tested using different combinations of ECG leads as inputs. The goal of this exercise was to determine the feasibility of using a reduced set of ECG leads in ECG interpretation and then reconstruct the missing ECG leads using a second model. Our results showed that the ECG interpretation model was able to maintain macro AUC score performance from 0.905 achieved with the full 12 leads down to 0.888 with as few as three leads, specifically leads I, II and V2. Using these results, three lead reconstruction models were developed to reconstruct 3, 4, or 5 missing chest leads. Of these three models, the best performance was achieved when reconstructing the 3 missing chest leads, reaching an average R2 score of 0.835 and root mean squared error of 0.265 mV. After developing and testing these models, the classification model was retrained with all available training data and quantized, post-training, using the LiteRT toolset to evaluate performance at lower precisions. Model performance was consistent across the macro AUC scores for the 32-bit model, float16 quantized model and dynamic range quantized model, maintaining a macro AUC of around 0.89 for the diagnostic output, 0.86 for the form output and 0.91 for the rhythm output for the 12-lead model. When 8-bit integer quantization was used, the macro AUC scores dropped substantially to around 0.5 for the 12-lead model
Fabrication and Characterization of Anionic Lipid-Containing Asymmetric Giant Unilamellar Vesicles in Physiological Ionic Strength Buffers
No full text**Abstract** This dissertation investigates the interactions of metal cations with anionic phospholipids, focusing on phosphatidylserine (PS) and phosphoinositides (PIPs), in the context of asymmetric giant unilamellar vesicle (aGUV) fabrication. By adapting a hemifusion-based method, we evaluated the effects of Ca²⁺ and Mg²⁺ on lipid bilayer composition and vesicle quality. Mg²⁺ was found to preserve homogeneity in supported lipid bilayers (SLBs) containing PI(4,5)P₂, unlike Ca²⁺, which induced micron-scale domain formation. Despite Mg²⁺ improving aGUV quality, broad lipid distributions remained in both leaflets. We developed a selection algorithm using fluorescence intensity metrics (Pex and Pa%) to identify aGUVs with near-theoretical leaflet compositions. Additionally, the thermal stability of Ca²⁺- and Mg²⁺-induced PIP domains was assessed, revealing Ca²⁺-dependent domains to be stable up to 80°C and further stabilized by cholesterol, while Mg²⁺ had negligible effects. These observations, interpreted through molecular dynamics simulations, support the relevance of the Hofmeister series and the Law of Matching Water Affinities in explaining PIP clustering behavior
Group Fairness in Ranking-Based Decisions: Do Metrics Reflect People’s Perception?
No full textAs automated decision-making systems increasingly influence critical areas such as hiring, university admissions, and bail decisions, concerns have grown over their potential to perpetuate biases and disproportionately harm marginalized groups. Ranking-based decision-making can amplify disparities due to reliance on historical data and predefined criteria, raising pressing questions about fairness, transparency, and accountability. While fairness metrics aim to evaluate and mitigate bias, a fundamental challenge remains: do these metrics align with People’s perceptions of fairness? This dissertation investigates this alignment through human-centered studies on group fairness, guiding the design of ranking systems that are both mathematically fair and widely accepted. We study two popular fairness metrics:: Attribute Rank Parity (ARP) and Normalized Discounted Cumulative KL-divergence (NDKL). To explore their alignment with People’s perceptions, we examine how these metrics relate to subjective fairness judgments, whether candidates’ demographic attributes (e.g., race) influences fairness perceptions, how fairness perceptions change when ranking candidates with similar versus dissimilar performance scores, and how list length (short vs. long) impacts fairness perceptions. We investigate: (RQ1) How ranking fairness metrics align with people’s subjective perceptions of fairness?, (RQ2) Does the inclusion of demographic information, such as race, influence people’s perception of these fairness metrics?, (RQ3) Do people’s perceptions of fairness change when candidates being ranked are similar in their score-based performance (such as grades of students) versus dissimilar?, and (RQ4) How does the number of ranked candidates influence people’s subjective perceptions of fairness? In Part I, we use a Likert-scale study with 480 participants to evaluate fairness perceptions across 12 ranking conditions, varying group sensitivity, group size, and performance scores. Results indicate a strong reliance on explicit score values, with participants favoring rankings that closely align with performancebased orderings. Group sensitivity played a role, as fairness perceptions shifted when rankings included demographic distinctions, particularly in unbalanced group sizes. Additionally, fairness perceptions were more consistent when candidates had similar grades, regardless of group sensitivity or size. In Part II, we address the limitations of Likert-scale studies, such as response ambiguity and scale interpretation differences, by introducing Just Noticeable Difference (JND) and two-alternative forced choice (2AFC) methodologies. In an experiment with 224 participants and 170,000 comparative judgments, we examined how ranking size (20 vs. 100 candidates) influenced fairness perceptions. Results revealed systematic differences in fairness metric alignment and unexpected cases where metrics diverged from People’s intuitions. Notably, longer lists led to more consistent and precise fairness judgments, suggesting that people make more stable fairness assessments when evaluating larger sets of ranked candidates. A thematic analysis of 6,000 crowdsourced responses revealed two key themes influencing fairness perceptions. Merit-based discontent emerged as participants expressed concerns about inconsistencies between grades and ranking orders, often voicing skepticism about fairness-driven ranking modifications. Transparency in socio-economic adjustments was another major concern, with participants emphasizing the need for clearer explanations when rankings incorporated group-based fairness interventions. Participants also used cognitive strategies (pattern detection, ratio comparisons) and visual cues (color distribution, symmetry, balance) to assess fairness, demonstrating that fairness perceptions are shaped not only by numerical fairness metrics but also by how rankings are visually structured and presented. Our findings highlight the challenges of operationalizing fairness in rankings, emphasizing the importance of refining fairness metrics to better align with People’s perceptions. These insights contribute to the design of fairness-aware ranking systems that balance quantitative fairness metrics with qualitative People’s intuitions. Future work should explore how transparency interventions influence fairness perceptions, refine perceptionaligned fairness metrics, and apply these insights to real-world ranking applications in hiring, education, and resource allocation
Climate and Environmental Oral History
No full textIn this project, we collected oral histories about climate change and its influence on daily life. We created open-ended questions using the TED (Tell, Explain, Describe) Method. We hope to spread awareness about climate change and its effects on Armenia, through stories of community members and expert knowledge. By combining personal narratives with contextual data, we aim to preserve these accounts as both a resource for climate research and a tool for teaching oral history
Orran's Vanadzor Land Use Assessment
No full textDuring this IQP, we created a development proposal to generate a sustainable revenue stream for Orran, a non-profit organization that supports underprivileged children and the elderly in Armenia. Through expert interviews, market research, and 3D modeling, we designed the proposed guesthouse and included supporting drawings to illustrate its layout and features. Various amenities and activities were also suggested, outlining the feasibility of each item and providing additional ways to improve Orran’s financial security
Expression, Purification, and Analysis of Several Variants of Renalase: a Potential Target for Type 1 Diabetes
No full textType 1 diabetes (T1D) is a disease caused by the autoimmune destruction of insulin-producing beta cells. A genome-wide CRISPR screen in T1D model mice conducted by the Joslin Diabetes Center found that deficiency in renalase protected beta cells against autoimmune destruction. Renalase is a largely uncharacterized enzyme with unknown function. The goal of the renalase project is to develop potent, specific small molecule inhibitors for renalase as a T1D treatment and to further understanding of the structure, substrates, and function of renalase as an enzyme. This MQP focused on the expression, purification, and analysis of several variants of renalase. Wild type, mutant 5, and mutant 6 human renalases and rat and bacteria renalases were expressed in E. coli cells and purified. Mutants 5 and 6 had significantly higher soluble expression in E. Coli than wild type. Several amines were identified as substrates for wild type human renalase. 4DHNAD was identified as a substrate for mutant 5 human renalase. A wild type human renalase and pargyline complex was crystallized