Physical Sciences and Mathematics Commons^™

Understanding Data Through The Lens Of Topology, Quang Truong

Dartmouth College Master’s Theses

Machine learning depends on the ability to learn insightful representations from data. Topology of data offers a rich source of information for constructing such representations, yet its potential remains under-explored by the broader machine learning community. This work investigates the power of applied topology through two complementary projects: Topological Message Passing with Path Complexes and Persistent Homology for Anomaly Detection. In the first project, we extend the topological message passing framework by introducing a novel approach centered on path complexes, where paths form the fundamental building blocks. Our theoretical analysis demonstrates that this model generalizes existing topological deep learning and …

Analysis Of Macular Pigment Carotenoids In Human Blood Serum Of Glaucoma Patients As A Measure Of Ocular Health: A Raman Spectroscopic Study., Joy Udensi, Ekaterina Loskutova, James Loughman, Hugh J. Byrne

Datasets

Carotenoids are a major component of the human diet and have been widely studied for their antioxidant, and vision protection roles in the human body, and dietary supplementation is promoted in particular for ocular health. An initial trial (European Nutrition in Glaucoma Management (ENIGMA)), which assessed macular pigment optical density (MPOD) as well as ocular structural, functional and perceptual parameters before and after the 18-month supplementation of glaucoma patients with macular pigment (MP) carotenoids (lutein, zeaxanthin, meso-zeaxanthin), confirmed supplementation significantly improved the clinical ocular health of participants. Blood contains all major dietary carotenoids, presenting it as a suitable and efficient …

Generalised Zero-Shot Learning For Action Recognition Fusing Text And Image Gans, Kaiqiang Huang, Susan Mckeever, Luis Miralles-Pechuán

Articles

Generalized Zero-Shot Action Recognition (GZSAR) is geared towards recognizing classes that the model has not been trained on, while still maintaining robust performance on the familiar, trained classes. This approach mitigates the need for an extensive amount of labeled training data and enhances the efficient utilization of available datasets. The main contribution of this paper is a novel approach for GZSAR that combines the power of two Generative Adversarial Networks (GANs). One GAN is responsible for generating embeddings from visual representations, while the other GAN focuses on generating embeddings from textual representations. These generated embeddings are fused, with the selection …

Wave Energy Converter Wave Force Prediction Using A Neural Network, Morgan Kline

Dissertations, Master's Theses and Master's Reports

Due to the unpredictable nature of large bodies of water, wave energy can be a difficult renewable resource to rely on. One way to make Wave Energy Converters (WECs) more efficient is to apply a control strategy. In many control solutions, it is assumed that the wave excitation force is known into the future. In many instances, especially with complex waveforms, this is simply not the case. Simulation studies have shown the promise of wave force prediction using neural networks. This study demonstrates this experimentally and aims to characterize the important factors when designing such a network. Several wave elevation …

Stigma And Efficacy Beliefs Regarding Opioid Use Disorder Treatment And Naloxone In Communities Participating In The Healing Communities Study Intervention, Nicky Lewis, Barry Eggleston, Redonna K. Chandler, Dawn Goddard-Eckrich, Jamie E. Luster, Dacia Beard, Emma Rodgers, Rouba A. Chahine, Philip M. Westgate, Shoshana N. Benjamin, Janae Holloway, Thomas Clarke, R. Craig Lefebvre, Michael D. Stein, Donald W. Helme, Jennifer Reynolds, Sharon L. Walsh, Darcy Freedman, Nabila El-Bassel, Kara Stephens, Anita Silwal, Michelle R. Lofwall, Janet E. Childerhose, Hilary L. Surratt, Brooke N. Crockett, Amy L. Farmer, James L. David, Laura Fanucchi, Judy Harness, Ben Wilburn, Kelli Bursey, Kristin Mattson, Sarah Mann, Rebecca D. Jackson, Aimee Shadwick, Katherine Calver, Deborah Chassler, Jennifer Kimball, Nancy Regan, Jeffrey H. Samet, Rachel Sword-Cruz, Michael D. Slater

Biostatistics Faculty Publications

Background The HEALing Communities Study (HCS) included health campaigns as part of a community-engaged intervention to reduce opioid-related overdose deaths in 67 highly impacted communities across Kentucky, Massachusetts, New York, and Ohio. Five campaigns were developed with community input to provide information on opioid use disorder (OUD) and overdose prevention, reduce stigma, and build demand for evidence-based practices (EBPs). An evaluation examined the recognition of campaign messages about naloxone and whether stigma and efficacy beliefs regarding OUD treatment and naloxone changed in HCS intervention communities.

Methods Data were collected through surveys offered on Facebook/Instagram to members of communities participating in …

Neural Correlates Of Audiovisual Narrative Speech Perception In Children And Adults On The Autism Spectrum: A Functional Magnetic Resonance Imaging Study, Lars A. Ross, Sophie Molholm, John S. Butler, Victor A. Del Bene, Tufikameni Brima, John J. Foxe

Articles

Autistic individuals show substantially reduced benefit from observing visual articulations during audiovisual speech perception, a multisensory integration deficit that is particularly relevant to social communication. This has mostly been studied using simple syllabic or word-level stimuli and it remains unclear how altered lower-level multisensory integration translates to the processing of more complex natural multisensory stimulus environments in autism. Here, functional neuroimaging was used to examine neural correlates of audiovisual gain (AV-gain) in 41 autistic individuals to those of 41 age-matched non-autistic controls when presented with a complex audiovisual narrative. Participants were presented with continuous narration of a story in auditory-alone, …

Analysing Natural Language Processing Techniques: A Comparative Study Of Nltk, Spacy, Bert, And Distilbert On Customer Query Datasets., Patrizia De Camillis

ICT

This study examines the role of sentiment analysis in customer queries, emphasising its impact on brand perception and the risks of poor query management. It compares the performance of NLP models—NLTK, spaCy, BERT, and DistilBERT—on customer query and feedback data. The findings show that BERT and DistilBERT produce similar results, often categorising queries as neutral, indicating their strength in handling diverse sentiments. NLTK and spaCy also share performance patterns. The research offers insights into the capabilities and limitations of these models in sentiment analysis.

Evaluating The Performance Of Different Long Short-Term Memory Networks (Lstm’S) On Financial Timeseries Data Using Mean Squared Error In Order To Identify The Optimum Lstm Variant For Regression Performance On Financial Timeseries Data., Patrick O’ Connor

ICT

This study explores the use of Long Short Term Memory (LSTM) networks, a variant of Recurrent Neural Networks (RNNs), in the context of financial forecasting, specifically oil price prediction. The research follows the CRISP-DM (Cross-Industry Standard Process for Data Mining) methodology and tests six different LSTM variants. The models are evaluated based on Mean Squared Error (MSE), aiming to determine the optimal parameter settings for each LSTM type. Among the variants tested, the Gated Recurrent Unit (GRU) emerged as the highest performer, achieving an MSE of 0.100. This was surprising, as simpler variants outperformed more complex ones, suggesting that simpler …

Ml Predictive Model For Earthquakes Integrating Mass, Distance, Gravity, And Magnitude., Aadarsh Kushwaha

ICT

This research investigates the application of machine learning regression models to improve earthquake prediction by integrating geophysical and astronomical factors such as Earth-Moon gravitational forces, their varying distances, and localized gravity fluctuations. Using data from 2011 to 2024, sourced from the US Geological Survey (USGS) and web scraping, the study tested models across four dataset proportions (25%, 50%, 80%, and 100%) with a 70:30 train-test split. The XGBRegressor model emerged as the best performer, achieving an R² score of 0.8706 on training data and 0.8632 on test data, along with a Mean Squared Error (MSE) of 0.1114 and Mean Absolute …

Maize Crop Pests And Diseases Classification Using Hybrid Models., Diana Flora Namaemba

ICT

This research focuses on improving the detection and classification of maize crop pests and diseases to enhance agricultural yield and food security. A dataset comprising 5389 images of maize conditions (healthy, pest-affected, and disease-affected) across seven classes was used. The images underwent preprocessing, including resizing to 299x299, class balancing using augmentation techniques, and noise reduction with Gaussian filtering.

Feature extraction utilised EfficientNetB0 and InceptionV3 architectures, with PCA employed for feature selection. Classification was conducted using a Support Vector Machine (SVM) with a One-vs-One strategy, alongside a baseline 2D CNN model. Data engineering included label encoding, standardisation, and an 80:10:10 train-test-validation …

Statistical And Machine Learning Techniques For Predicting Solar Power Generation In A Microgrid., Conor Dillon

ICT

This study investigates statistical and machine learning models for forecasting solar power generation in microgrids, focusing on the solar installation at Powell-Focht Bioengineering Hall, UC San Diego. Accurate predictions are critical due to the variability of solar energy, aiming to optimise microgrid operations and solar power efficiency. The research compares the performance of SARIMAX, LSTM, Random Forest, and ANN models using meteorological and solar power time series data. It finds that current meteorological inputs, especially solar radiation, enhance short-term forecasting accuracy over reliance on historical patterns.

The Random Forest Auto Regressor (RFAR) outperformed other models in 10-day-ahead solar power forecasting, …

Supply Chain Optimisation With Machine Learning And Neural Networks: Applications To Demand Planning, Supply Planning, And Inventory Planning., Laurence Cully

ICT

This thesis explores the impact of machine learning (ML) on supply chain planning, particularly in demand forecasting, supply planning, and inventory optimisation. By analysing literature on supply chain management, data flow, and the intersection of ML and competitive advantage, the author contextualises the research within a globalised market's demands. Case studies, interviews with industry professionals, and raw data collection provide empirical support for evaluating the research objectives and documenting the integration of ML in supply chain processes.

The findings reveal that optimised ML models, particularly those using model stacking (autoregressors, GRUs, and Random Forests), significantly outperform traditional demand forecasting methods, …

Random Walk Methods For Geometry Representation Agnostic Transport, Dario R. Seyb

Dartmouth College Ph.D Dissertations

In computer graphics, we use geometry representations to model a wide range of virtual scenes—from the fantastical worlds shown in animated movies to intricate mechanical parts.
These representations provide the context for transport problems—light transport is used to produce images of virtual scenes and diffusive transport to simulate distributions of quantities like heat.

There are many types of representations each with their own advantages.
For example, explicit ones make it easy to directly manipulate surfaces, while implicit representations allow for intuitive modeling by non-technical users and straightforward integration into machine learning systems.
Unfortunately, many algorithms that work on these digital …

Halide-Assisted Growth Of Transition Metal Dichalcogenides, Vinaayak Sivam Balasubramaniam

Dissertations, Master's Theses and Master's Reports

Monolayers of transition metal dichalcogenides (TMDCs) have attracted significant attention as the rare two-dimensional (2D) semiconducting materials with a direct energy band gap. Chemical vapour deposition (CVD) is one of the scalable techniques to grow atomically thin TMDC monolayers in high quality, but it requires high growth temperature. Herein we report the growth of MoS2, WS2 and MoSe2 by a one-step halide-assisted CVD method using NaCl and KCl as the catalysts. These halides could reduce the growth temperature of TMDCS by reacting with the precursors (TMDC powders) to form volatile intermediate compounds as the growth species. We use optical microscopy, …

Statically Controlled Synchronized Lane Architectures, Scott K. Pomerville

Dissertations, Master's Theses and Master's Reports

Modern superscalar processors dominate the field of computing. While dynamic execution allows for versatility in code, these processors are complex. Statically scheduled code has historically enabled simpler processor designs, but static scheduling cannot account for variables that are unknown at compile time. Furthermore, static scheduling has many inefficiencies, such as the need to insert a large number of nops for code in traditional Very Long Instruction Word (VLIW) processors. In this dissertation, we explore a novel architectural approach for statically scheduled code by breaking the code into several synchronous instruction streams. By representing code in a fundamentally new way, we …

A Physicochemical Characterization Of Protein Aggregates: Influence Of Lipid Membranes And Surfaces, Connor J. Hall

Dissertations, Master's Theses and Master's Reports

Protein aggregation is a hallmark of neurodegenerative diseases (NDs). Although there has been a greater emphasis in the past decade or more on the correlation between the physicochemical characteristics of aggregates and their cellular toxicity, it is still unclear what specific features of an aggregate render it toxic. To better understand that, there has been a renewed focus on the characterization of unique biochemical and biophysical properties of protein aggregates. Elucidation of these unique characteristics can help to better understand disease pathogenesis in vivo. Furthermore, the lipid-rich nature of the brain sets it apart from all other organs, and numerous …

Enhancing Students’ User Experience With A Code Critiquer, Laura E. Albrant

Dissertations, Master's Theses and Master's Reports

This thesis explores the role of human factors in the realm of code critiquers and students’ experiences with them. Across three studies, the work utilized Design Thinking to improve the user experience of WebTA for introductory engineering students learning MATLAB. The first two studies gathered observational and interview data to empathize, define, and ideate a new user interface (UI). Said UI was prototyped and then tested with the third study. Overall, the surveys’ data suggests that most students found the new design to be more appealing, useful, and purposeful; however, there is still plenty of room for improvement. Additionally, analysis …

Constructing The 4th Hawc Catalog Of Very-High-Energy Sources Using An Automated Likelihood Pipeline Search, Samuel J. Groetsch

Dissertations, Master's Theses and Master's Reports

In this thesis, I present an analysis pipeline that allows one to search for and identify sources of very-high-energy (VHE) γ-ray emission in sky maps created from data collected with the High Altitude Water-Cherenkov (HAWC) Observatory. The HAWC Observatory is located at about 4100 m elevation just below Pico de Orizaba, the highest mountain in Mexico also known as Citlaltepetl (from the Nahuatl words citlal(in) = star, and tepetl = mountain). The observatory is an array of several hundred detector stations that measure signals from particle showers, so-called extensive air showers (EAS), that are initiated in interactions of primary cosmic …

Governing Dynamic Watersheds Under Static Institutions – Examining How Local Governments Respond To Floods And Droughts In The United States, Jonathon R. Loos

Dartmouth College Ph.D Dissertations

The U.S. has seen a shift towards decentralized watershed governance in recent decades that has increased the delegation of management responsibilities to local governments and community organizations. This shift has precipitated the emergence of multilevel watershed governance systems (e.g. national, state, regional, local management levels) that are hypothesized to be more adaptive and responsive to local needs. However, multilevel governance systems risk complicating and overburdening the role of local governments within watershed management, and little is known about how local governments address socio-ecological change within multilevel institutions. Working within several U.S. watershed geographies, this dissertation seeks to interrogate theories of …

An Investigation Into The Role Of Machine Learning And Deep Learning Models As A Means Of Leveraging The Ever-Expanding Volume Of Astronomical Data To Automate Stellar Classification., Gerard Heraghty

ICT

This research investigates the use of machine learning and neural network models for automated stellar classification in large astronomical surveys, addressing challenges posed by the increasing volume of data. Using the MK scheme as the classification standard, the study focused on spectroscopic data and balanced the dataset using SMOTE techniques to handle class imbalances. Various models, including Random Forest, SVM, MLP, and CNN, were trained and compared for classifying MK main and sub-classes. CNN achieved the highest accuracy (93.86%) for main class classification, while SVM excelled at sub-class classification (63.23%) on balanced datasets. However, when tested on real-world SDSS data, …

Leveraging Uncertainty To Understand The Causes And Consequences Of Observed Climate Change, Alexander Reid Gottlieb

Dartmouth College Ph.D Dissertations

Detecting and attributing climate change to date is essential to understanding the harm global warming has caused and the threat it poses in the future. Yet detecting and attributing historical climate change and its attendant impacts is challenging owing to observational uncertainties in the historical record, the difficulty of isolating the forced response from unforced variability, and the methodological ambiguity on how to best map climate variability and change onto climate impacts. In this thesis, I use the study system of Northern Hemisphere snow to demonstrate how synthesizing a wide array of observational and climate model data with methods from …

Plant Root Hair Adhesion Mediated Through Variation In Sodium Ion Concentration, Natasha Mulenga

Honors Program Theses

This study investigates the effects of sodium ion (Na+) concentration on the adhesion of Brassica Rapa root hairs using atomic force microscopy (AFM). Plant root hair adhesion to soil particles plays a crucial role in anchoring plants and facilitating nutrient uptake. Previous research has shown that calcium ions (Ca2+) cross-link pectin molecules in the cell wall, increasing rigidity and reducing adhesion. As Na+ binds to pectin without cross-linking, we expected that root hairs exposed to sodium chloride (NaCl) solutions would exhibit longer adhesive interactions and similar binding forces compared to those exposed to calcium chloride (CaCl2) solutions of equal concentrations. …

Transient Diode Laser Absorption Spectroscopy For Measuring Collisional Quenching Rates Of Co And Its Potential For Studying H2o, Jordan M. Polvere

Master’s Theses

Motivated by the necessity of non-LTE models that account for species that are not in Local Thermodynamic Equilibrium (LTE) in Titan’s and Earth’s atmospheres, collisional quenching rates of CO and H2O_(g) have been studied. Providing more accurate and precise values of collisional quenching rate coefficients is crucial in the development of more advanced atmospheric models, which are important to fields of science such as meteorology and astronomy. Rates of collisional quenching of CO(v=1,2) by CO and CO(v=2) by N₂ at approximately 301-321 K were measured by transient diode laser absorption spectroscopy. The rate coefficients of CO(v=1)-CO, CO(v=2)-CO, and …

Pitching The Use Of Squared And Interaction Terms In Regression Via Baseball Heat Maps, Lucas Chepelsky

Williams Honors College, Honors Research Projects

This project will examine the impact of using second-order terms in regression. For illustration, we use an example of regression where a baseball player's three by three heat map, including the height and distance from inside to outside of the pitch, are variables used to predict batting average. We find that second-order terms are crucial in discovering nonlinear relationships and interaction effects in regression models, and maintain that the common practice of using first-order additive models is insufficient.

Machine Learning-Based Analysis Of Dna Methylation Patterns In E.Lenta, Tsion Tsegaye Sherbeza

All Graduate Theses, Dissertations, and Other Capstone Projects

Methylation patterns in bacterial genomes, such as those found in Eggerthella lenta, take roles in mediating microbial interactions with their environment, host, and external stressors. These patterns are formed by DNA methylation with diverse sequence specificities that provide insights into the DNA sequence regulation and defense against bacteriophages. We utilize computational approaches and machine learning models to identify and analyze 5mC and 6mA methylation motifs in E. lenta genomes. By integrating host characteristics such as age, birth country, gender, and medication history, we explore 1) the predictive relationships between 5mC & 6mA methylation types in E. lenta strains and …

“Zero” Porosity High Loading Nmc622 Positive Electrodes For Li-Ion Batteries, Haidar Y. Alolaywi, Kubra Uzun, Yang-Tse Cheng

Chemical and Materials Engineering Faculty Publications

LiNi0.6 Mn0.2Co0.2 O 2 (NMC622) is a widely used positive electrode material for lithium-ion batteries, including electric vehicles. In this work, we investigated the effects of porosity, ranging from “zero” to the typical 35%, on the electrochemical behavior of high- loading NMC622 electrodes. Although it is well known that the energy density of the electrode increases with increasing areal capacity and decreasing porosity, NMC-positive electrodes with exceedingly low porosity (e.g., near zero) and high loading (e.g., 4 mAh cm−2 ) have not been investigated. Here, we report an intriguing observation that the “zero porosity” NMC electrode can have higher capacity …

The Educational Affordances And Challenges Of Chatgpt: State Of The Field, Helen Crompton, Diane Burke

STEMPS Faculty Publications

ChatGPT was released to the public in November 30, 2022. This study examines how ChatGPT can be used by educators and students to promote learning and what are the challenges and limitations. This study is unique in providing one of the first systematic reviews using peer review studies to provide an early examination of the field. Using PRISMA principles, 44 articles were selected for review. Grounded coding was then used to reveal trends in the data. The findings show that educators can use ChatGPT for teaching support, task automation, and professional development. These were further delineated further by axial sub …

Multiple Imputation For Robust Cluster Analysis To Address Missingness In Medical Data, Arnold Harder, Gayla R. Olbricht, Godwin Ekuma, Daniel B. Hier, Tayo Obafemi-Ajayi

Mathematics and Statistics Faculty Research & Creative Works

Cluster Analysis Has Been Applied To A Wide Range Of Problems As An Exploratory Tool To Enhance Knowledge Discovery. Clustering Aids Disease Subtyping, I.e. Identifying Homogeneous Patient Subgroups, In Medical Data. Missing Data Is A Common Problem In Medical Research And Could Bias Clustering Results If Not Properly Handled. Yet, Multiple Imputation Has Been Under-Utilized To Address Missingness, When Clustering Medical Data. Its Limited Integration In Clustering Of Medical Data, Despite The Known Advantages And Benefits Of Multiple Imputation, Could Be Attributed To Many Factors. This Includes Methodological Complexity, Difficulties In Pooling Results To Obtain A Consensus Clustering, Uncertainty Regarding …

A Case Study On Variations In Network Structure And Cross- Sector Alignment In Two Local Systems Serving Pregnant And Parenting Women In Recovery, Liza M. Creel, Yana Feygin, Madeline Shipley, Deborah Winders Davis, Tiffany Cole Hall, Chaly Downs, Stephanie Hoskins, Natalie Pasquenza, Scott D. Duncan

Biostatistics Faculty Publications

Objective: To describe network structure and alignment across organizations in healthcare, public health, and social services sectors that serve pregnant and parenting women with substance use disorder (SUD) in an urban and a rural community.

Data Sources and Study Settings: Two community networks, one urban and one rural with each including a residential substance use treatment program, in Kentucky during 2021.

Study Design: Social network analysis measured system collaboration and cross-sector alignment between healthcare, public health, and social services organizations, applying the Framework for Aligning Sectors. To understand the alignment and structure of each network, we measured network density overall …

Higher First 30-Day Dose Of Buprenorphine For Opioid Use Disorder Treatment Is Associated With Decreased Mortality, Feitong Lei, Michelle R. Lofwall, Jana Mcanich, Reuben Adatorwovor, Emily Slade, Patricia R. Freeman, Daniela Moga, Nabarun Dasgupta, Sharon L. Walsh, Rachel Vickers-Smith, Svetla Slavova

Biostatistics Faculty Publications

Objective: Buprenorphine is a medication for opioid use disorder that reduces mortality. This study aims to investigate the less well-understood relationship between the dose in the early stages of treatment and the subsequent risk of death.

Methods: We used Kentucky prescription monitoring data to identify adult Kentucky residents initiating transmucosal buprenorphine medication for opioid use disorder (January 2017 to November 2019). Average daily buprenorphine dose for days covered in the first 30 days of treatment was categorized as ≤8 mg, >8 to ≤16 mg, and >16 mg. Patients were followed for 365 days after the first 30 days of buprenorphine …