Physical Sciences and Mathematics Commons^™

Stereotypes And Language Models: Understanding How Language Models Encode Stereotypes, Debiasing Language Models, And Examining How Stereotypes Affect Conversations, Brian C. Wang

Computer Science Senior Theses

This thesis describes a variety of approaches in examining how language models encode stereotypes (understanding stereotypes from a model point-of-view), debiasing language models, and using language models to understand how stereotypes affect conversations (understanding stereotypes from a conversational point-of-view). We present a novel approach for textual clues analysis that makes language models more interpretable, combining the understanding of what stereotypes the internal structures of language models have encoded during their initial training (via attention-based analysis) and understanding what textual clues are most relevant to identifying stereotypes for models trained to detect stereotypes (via SHAP-based analysis). We find that different pre-trained …

Sarcasm Detection In English And Arabic Tweets Using Transformer Models, Rishik Lad

Computer Science Senior Theses

This thesis describes our approach toward the detection of sarcasm and its various types in English and Arabic Tweets through methods in deep learning. There are five problems we attempted: (1) detection of sarcasm in English Tweets, (2) detection of sarcasm in Arabic Tweets, (3) determining the type of sarcastic speech subcategory for English Tweets, (4) determining which of two semantically equivalent English Tweets is sarcastic, and (5) determining which of two semantically equivalent Arabic Tweets is sarcastic. All tasks were framed as classification problems, and our contributions are threefold: (a) we developed an English binary classifier system with RoBERTa, …

Counterfactual Replacement Analysis For Interpretation Of Blackbox Sexism Classification Models, Anders Knospe

Computer Science Senior Theses

This paper describes the AKD team’s system designed for SemEval-2023 Task 10: Explainable Detection of Online Sexism (EDOS). We implement a simple fine-tuned GPT-3 model, ranking 26 on the leaderboard for task A. We also discuss different approaches to interpretability in the context of critiquing the EDOS task’s sub-category oriented approach. Finally, we propose counterfactual replacement analysis, a novel prototype technique for approaching explainability.

Jones Polynomial Obstructions For Positivity Of Knots, Lizzie Buchanan

Dartmouth College Ph.D Dissertations

The fundamental problem in knot theory is distinguishing one knot from another. We accomplish this by looking at knot invariants. One such invariant is positivity. A knot is positive if it has a diagram in which all crossings are positive. A knot is almost-positive if it does not have a diagram where all crossings are positive, but it does have a diagram in which all but one crossings are positive. Given a knot with an almost-positive diagram, it is in general very hard to determine whether it might also have a positive diagram. This work provides positivity obstructions for three …

Utilizing Mixed Graphical Network Models To Explore Parent Psychological Symptoms And Their Centrality To Parent Mental Health In Households With High Child Screen Usage, Piper F. Stacey, Nicholas C. Jacobson, Damien Lekkas

Computer Science Senior Theses

Especially among adolescents, screens are being used more than ever. In conjunction with this trend, mental illness is increasingly prevalent among both adults and children, and parental psychological problems are shown to be associated with children's TV watching, video watching, and gaming (Pulkki-Råback et al., 2022). This study aims to approach parent mental illness symptom by symptom to explore which specific symptoms are most central to parent psychological problems in households where children show high screen time behaviors. We draw from the Adolescent Brain Cognitive Development Study (ABCD Study®), a nationwide sample of 11,875 children aged 10-13 collected by …

Data-Optimized Spatial Field Predictions For Robotic Adaptive Sampling: A Gaussian Process Approach, Zachary Nathan

Computer Science Senior Theses

We introduce a framework that combines Gaussian Process models, robotic sensor measurements, and sampling data to predict spatial fields. In this context, a spatial field refers to the distribution of a variable throughout a specific area, such as temperature or pH variations over the surface of a lake. Whereas existing methods tend to analyze only the particular field(s) of interest, our approach optimizes predictions through the effective use of all available data. We validated our framework on several datasets, showing that errors can decline by up to two-thirds through the inclusion of additional colocated measurements. In support of adaptive sampling, …

Exploring Improvements To Space-Bounded Derandomization From Better Pseudorandom Generators, Boxian Wang

Computer Science Senior Theses

Saks and Zhou used Nisan’s PRG in a recursive manner to obtain BPL ⊆ L^(3/2). We describe how this framework could be generalized to use arbitrary PRGs following Armoni’s sampler idea. We then give a theorem relating the seed length of a better PRG to the implied improvements in derandomizing BPL. Recently, Hoza used Armoni’s PRG in the Saks-Zhou framework to obtain an even better derandomization. We describe the construction of Armoni’s PRG and conjecture that by using basic components other than extractors, parameters in that construction could be improved. Under some assumptions, we calculate the extent to which such …

Deep Learning For Skin Photoaging, Gokul Srinivasan

Computer Science Senior Theses

Skin photoaging is the premature aging of skin that results from ultraviolet light exposure. It is a major risk factor for the development of skin cancer, among other malignant skin pathologies. Accordingly, understanding its etiology is important for both preventative and reparative clinical action. In this study, skin samples obtained from patients with ranging solar elastosis grades – a proxy for skin photoaging – were sequenced using next-generation sequencing techniques to further understand the genomic, epigenomic, and histological signs and signals of skin photoaging. The results of this study suggest that tissues with severe photoaging exhibit increases in the frequency …

Interpreting Business Strategy And Market Dynamics: A Multi-Method Ai Approach, Lobna Jbeniani

Computer Science Senior Theses

This research paper presents an integrated approach that combines Long Short-Term Memory (LSTM), Q-Learning, Monte Carlo methods, and Text-to-Text Transfer Transformer (T5) to analyze and evaluate the business strategies of public companies. Leveraging a large and diverse dataset sourced from multiple reliable sources, the study examines corporate strategies and their impact on market dynamics. LSTM and Q-Learning are employed to process sequential data, enabling informed decision-making in simulated market environments and providing insights into potential outcomes of different strategies. The Monte Carlo method manages uncertainty, allowing for a comprehensive analysis of risks and rewards associated with specific strategies. T5 interprets …

Unmasking Bias: Investigating Strategies For Minimizing Discrimination In Ai Models, Julia L. Martin

Computer Science Senior Theses

Artificial Intelligence (AI) models are increasingly used as predictive tools with real-world applications occurring in diverse fields ranging from the healthcare industry to the criminal justice system. While AI often offers efficient and relatively effective solutions, there are growing concerns regarding AI’s role in decision-making processes due to potential biases embedded in these models. In many cases, bias in AI models can produce unfair outcomes, perpetuate social inequities, and undermine the trustworthiness of AI systems. This thesis explores this problem and spotlights certain biased models that are currently utilized in real-world situations. One such example is a highly biased AI …

Connecting Linguistic Expressions And Pain Relief Through Transformer Model Construction And Analysis, Sarah M. Chacko

Computer Science Senior Theses

Chronic pain is a widespread problem that significantly impacts quality of life. Overprescription and abuse of pain medication continues to be a major public health issue and can further burden patients due to a fragmented health care system. Previous research has suggested a possible psychological basis to pain and the potential for safer, non-pharmacological alternatives for pain relief. This project leverages language models to study chronic pain development and relief through psychological treatments, which will be assessed through responses to post-treatment interviews. A transformer-based natural language processing model is employed to identify connections between language expressions and pain on a …

An Algorithmic Approach To Jazz Guitar Voice-Leading Chord Fingerings, Matthew B. Keating

Computer Science Senior Theses

A problem in guitar practice is choosing chord voicings that fit together in sequence, a process known as voice leading. In jazz, a guitarist follows voice leading by maintaining stepwise or limited motion for smoother harmony. The main avenues to learn jazz guitar voice leading theory are through a guitar instructor or chord books. To our knowledge, no computational method of generating voice-leading given chord labels exists. First, we demonstrate the complexity of this problem by presenting a graph search algorithm to optimize for a simplified version of voice leading. Then, we present a novel approach to algorithmically derive tablature …

Investigating English-Language Dialect-Adjusted Models, Samiha Datta

Computer Science Senior Theses

This thesis describes several approaches to better understand how large language models interpret different dialects of the English language. Our goal is to consider multiple contexts of textual data and to analyze how English-language dialects are realized in them, as well as how a variety of machine learning techniques handle these differences. We focus on two genres of text data: news and social media. In the news context, we establish a dataset covering news articles from five countries and four US states and consider language modeling analysis, topic and sentiment distributions, and manual analysis before performing nine experiments and evaluating …

Utilizing Natural Language Processing For Automated Clinical Text Review: Identification Of Care Preference Documentation In Patients’ Discharge Summaries, Saksham Arora

Computer Science Senior Theses

Improving patient-centered care necessitates accurate documentation of care preferences, a crucial aspect often underrepresented in administrative data. Most studies apply care documentation to specific patient populations, rather than more appropriately broad population of `seriously ill' patients. This paper addresses this gap by leveraging transformer-based machine learning models, exhibiting an improvement over traditional keyword-based search methods in identifying care preference documentation.

In order to capture a broad spectrum of seriously ill patients, we matched decedent patients to non-decedent counterparts by utilizing a propensity score matching, accounting for important variables like age, gender, primary diagnoses and commodities. We trained and fine-tuned Bio_ClinicalBERT …

Georcf-Gn: Geography-Aware State Prediction In Dynamic Networks, Barkin Cavdaroglu

Computer Science Senior Theses

No abstract provided.

Effective Non-Hermiticity And Topology In Markovian Quadratic Bosonic Dynamics, Vincent Paul Flynn

Dartmouth College Ph.D Dissertations

Recently, there has been an explosion of interest in re-imagining many-body quantum phenomena beyond equilibrium. One such effort has extended the symmetry-protected topological (SPT) phase classification of non-interacting fermions to driven and dissipative settings, uncovering novel topological phenomena that are not known to exist in equilibrium which may have wide-ranging applications in quantum science. Similar physics in non-interacting bosonic systems has remained elusive. Even at equilibrium, an "effective non-Hermiticity" intrinsic to bosonic Hamiltonians poses theoretical challenges. While this non-Hermiticity has been acknowledged, its implications have not been explored in-depth. Beyond this dynamical peculiarity, major roadblocks have arisen in the search …

Expressive Marks: Art In The Age Of Augmented Reality, Carson G. Levine

Dartmouth College Master’s Theses

Augmented reality (AR) and non-fungible tokens (NFTs) introduce new considerations for the long-standing debate of what it means for digital art to be “real.” However, the ability to create AR experiences is limited to those who are technically skilled or who can afford to consult someone else. This paper addresses the need for an accessible tool that enables artists of all technical backgrounds to expressively create marks in AR. The solution includes a mobile application called CrayonAR. The system was designed to be modular, minimal, and physically engaging, and was developed in Unity using ARFoundation and Firebase Storage and Realtime …

Downstream Gradients In Unit Stream Power Influence Log Jam Location And Process Domain, Eliza H. Malakoff

Dartmouth College Master’s Theses

Growing calls for the use of natural materials and processes to meet management goals have positioned artificial log jams as a compelling alternative to hard engineering instream and floodplain habitat. Deep uncertainties remain, however, about where and how wood should be placed to best mimic natural river processes. In this study, I test whether at-a-point or downstream gradients in unit stream power, an estimate of a river’s ability to do work, exert control over where and how log jams form. Using field observations of 360 log jams in New Hampshire and Vermont and an additional 320 previously published locations of …

Sprout: Using A Garden Metaphor To Visualize And Support Customizable And Collaborative Health Tracking, Pape Sow Traoré

Dartmouth College Master’s Theses

Self-tracking tools have become increasingly popular, especially with the advent of wearable technology and smartphone applications. However, traditional tracking tools often display data in a quantitative format that can be overwhelming and cause users to abandon their tracking efforts. Additionally, these tools typically provide a generic user experience and are designed from a single-user perspective, lacking external support. To overcome these limitations, we develop Sprout, a mobile data-tracking application that offers a more qualitative, customizable, and collaborative experience for health monitoring and management. Sprout uses a garden metaphor to visually represent health information and allows users to tailor their …

Brill--Noether Theory Via K3 Surfaces, Richard Haburcak

Dartmouth College Ph.D Dissertations

Brill--Noether theory studies the different projective embeddings that an algebraic curve admits. For a curve with a given projective embedding, we study the question of what other projective embeddings the curve can admit. Our techniques use curves on K3 surfaces. Lazarsfeld's proof of the Gieseker--Petri theorem solidified the role of K3 surfaces in the Brill--Noether theory of curves. In this thesis, we further the study of the Brill--Noether theory of curves on K3 surfaces.

We prove results concerning lifting line bundles from curves to K3 surfaces. Via an analysis of the stability of Lazarsfeld--Mukai bundles, we deduce a bounded version …

New Physics In The Age Of Precision Cosmology, Vivian I. Sabla

Dartmouth College Ph.D Dissertations

The Lambda-cold dark matter (LCDM) model has become the standard model of cosmology because of its ability to reproduce a vast array of cosmological observations, from the earliest moments of our Universe, to the current period of accelerated expansion, which it does with great accuracy. However, the success of this model only distracts from its inherent flaws and ambiguities. LCDM is purely phenomenological, providing no physical explanation for the nature of dark matter, responsible for the formation and evolution of large-scale structure, and giving an inconclusive explanation for dark energy, which drives the current period of accelerated expansion.

Furthermore, cracks …

Synthetic, Catalytic, And Mechanistic Studies Of Supermesityl Phosphiranes And Phosphines, Ryan M. Tipker

Dartmouth College Ph.D Dissertations

Methylation of P-stereogenic phosphiranes Mes*PCH2CH(R) (Mes* = 2,4,6-(t-Bu)3C6H2, R = Me, Ph) with MeOTf gave P-stereogenic phosphiranium cations; [Mes*P(Me)CH2CH(Ph)][OTf] underwent syn-anti isomerization via P-epimerization. Mechanistic studies suggested ring opening gave a hyperconjugation-stabilized carbocation in which pyramidal inversion at P was promoted by s-interaction with the pendant cation. Attempted phosphirane protonation with HOTf resulted in ring opening and C-H activation of an o-t-Bu group to give phospholanium cations. Treatment of [Mes*P(Me)CH2CH(Ph)][OTf] with LiPPh2 gave bis(phosphino)ethanes. Copper- catalyzed P-alkylation of the secondary phosphine PHPh(Mes*) with benzyl bromides gave P-stereogenic tertiary phosphines with a supermesityl substituent.

Beyond News Values On Twitter: Predicting Factors That Drive User Engagement In News, Zhiyan Zhong

Dartmouth College Master’s Theses

When deciding on what news stories to cover, traditional journalism determines news values by following several elements of newsworthiness, such as impact, timeliness, and prominence. However, these guidelines do not always seem to correspond with the success of content on social media. As people are increasingly turning to social media for news, our research aims to understand and predict factors that drive user engagement for news on social media. In this study, we analyze news content published on Twitter, and examine a diverse set of characteristics like metrics retrieved from the Twitter API and semantics by natural language processing, including …

The Extremes Of Galaxy Formation & Evolution, Kelly E. Whalen

Dartmouth College Ph.D Dissertations

Galaxy populations are shaped by the physical processes that regulate their star formation and central black hole growth throughout cosmic time. The primary aim of this thesis is to understand how these processes occur and how they shape evolution in some of the most extreme galaxies in the Universe including quasars, compact starbursts, and ultra-diffuse dwarfs. Gas-rich major mergers funnel large amounts of gas towards the nucleus, triggering rapid AGN accretion and compact star formation. In this work, I study powerful quasars and extreme, massive, compact starburst galaxies within the context of merger-driven galaxy evolution scenarios. One aim of this …

Counting Elliptic Curves With A Cyclic M-Isogeny Over Q, Grant S. Molnar

Dartmouth College Ph.D Dissertations

Using methods from analytic number theory, for m > 5 and for m = 4, we obtain asymptotics with power-saving error terms for counts of elliptic curves with a cyclic m-isogeny up to quadratic twist over the rational numbers. For m > 5, we then apply a Tauberian theorem to achieve asymptotics with power saving error for counts of elliptic curves with a cyclic m-isogeny up to isomorphism over the rational numbers.

An Empirical Study Of Locality-Sensitive Hashing To Approximate The Minimum Spanning Tree, Elizabeth Crocker

Computer Science Senior Theses

The minimum spanning tree is a problem with important applications but for which there are no known efficient algorithms for large data sets. Locality-sensitive hashing has been used to solve the near-neighbor problem and further applications in clustering, which indicates its potential for approximating the minimum spanning tree as well. An algorithm by Sariel Har-Peled, Piotr Indyk, and Rajeev Motwani utilizes locality-sensitive hashing to provide a c-approximation of the minimum spanning tree in O(dn^1+1/c log² n) time. In this thesis, we implement and test this algorithm. We determine that the algorithm is suited to provide a better-than-random approximation …

Cyclic Mixed-Radix Dense Gray Codes, Jessica Cheng

Computer Science Senior Theses

A Gray code is a sequence of n binary integers in the range 0 to n-1 that has the Gray-code property: each integer in the sequence differs from the integer before it in a single digit. Gray codes have many applications, ranging from rotary encoders to Boolean circuit minimization. We refer to Gray codes where the first and last
codewords in the sequence fulfill the Gray-code property as cyclic. Additionally, we refer to a Gray code as dense if the sequence of n numbers consists of a permutation of ⟨0, 1, . . . , n − 1⟩. This thesis …

Integrating Remote And In-Situ Techniques To Quantify Landscape Evolution, Matthew Maclay

Dartmouth College Master’s Theses

With the increasing availability and resolution of remote sensing techniques, the resulting data products are increasingly being applied to answer societally relevant questions regarding quantifying the effects of climate change, mitigating natural hazards, and understanding landscape changes over varying temporal and spatial scales. While the power and potential for such large-scale, efficient, and cost-effective surveys are undeniable, a thorough understanding of any environment requires that remotely sensed data are ground-truthed or put into context with in-situ observations. In this thesis, Chapter 1 presents a literature review of Martian analog sites and discusses the importance of integrating in-situ and remote sensing …

Spectral Sequences And Khovanov Homology, Zachary J. Winkeler

Dartmouth College Ph.D Dissertations

In this thesis, we will focus on two main topics; the common thread between both will be the existence of spectral sequences relating Khovanov homology to other knot invariants. Our first topic is an invariant MKh(L) for links in thickened disks with multiple punctures. This invariant is different from but inspired by both the Asaeda-Pryzytycki-Sikora (APS) homology and its specialization to links in the solid torus. Our theory will be constructed from a Z^n-filtration on the Khovanov complex, and as a result we will get various spectral sequences relating MKh(L) to Kh(L), AKh(L), and APS(L). Our …

Wildfire Activity, Climate Response, And Ice Core Signal Preservation In The North Pacific Region, Margaret Lonergan

Dartmouth College Master’s Theses

Wildfires have become more destructive over recent decades with climate change, so understanding how fire regimes will change with further climate change is critical for effective fire management practices. Paleofire records provide insight into how fire regimes have responded to temperature and precipitation variability in the past. Ice cores, such as the Denali ice core from central Alaska, capture regional-scale fire proxies including black carbon at an annual resolution for centuries to millennia. This makes them ideally suited to construct high temporal resolution, regional paleofire records extending back into the Common Era. However, it is critical to understand the instrumental …