Artificial Intelligence and Robotics Commons^™

Biologically Informed Negative Samplingfor Antibody Chain Pairing Classification, Ishita Singh

Computer Science Senior Theses

Antibody heavy and light chain (H/L) pairing is fundamental to antigen recognition and stability. While single-cell sequencing preserves native pairing information, widely used bulk repertoire and spatial transcriptomics platforms do not, motivating the need for efficient ML methods to infer H/L pairing. Training a binary classifier for this task faces the methodological challenge of a lack of true biological negatives, since natural selection eliminates B cells with incompatible H/L pairs.

In this thesis, I introduce a biologically informed negative sampling strategy for H/L pairing classification, drawing on known V-gene biases in heavy and light chain pairing. Pseudo-negatives are constructed by …

Capturing Large Language Model Similarity Through Spectral Analysis, Ishan Verma Prasad

Computer Science Senior Theses

With the rapid development of open-sourced models on Huggingface, there is a strong need for a way to systematically determine the similarity between models. More strongly, for intellectual property and organization, we need a way to determine the "lineage" of models. We borrow principles from Heavy-Tailed Self-Regularization and Random Matrix Theory to provide an inference-free method to accomplish this. We cluster a corpus of several model families by their spectral fingerprints and demonstrate that each model family occupies a distinct region in weight space. This confirms prior ideas of training setups leaving artifacts on model weights and allows us to …

High-Throughput Robotic Ethanol Inhibition Assays For Engineered Thermophilic Biofuel Strains, Kevin He, Daniel Olson, Marybeth Maloney, Anthony Lanahan

Wetterhahn Science Symposium Posters

Ethanol stress assays are commonly used to evaluate microbial tolerance, metabolic adaptation, and fermentation performance. However, manual liquid handling introduces variability across replicate wells and small-volume pipetting steps, limiting reproducibility and throughput. This study developed an automated OT-2 robotic workflow to generate replicated ethanol concentration gradients for high-throughput inhibition assays in engineered thermophilic biofuel strains. Kinetic plate-reader measurements were used to quantify ethanol-dependent growth responses under anaerobic fermentation conditions. The reasearch question is: How do engineered thermophilic biofuel strains differ in ethanol-dependent growth inhibition under anaerobic fermentation conditions, and can automated robotic assays improve the reproducibility of these measurements? Can …

Storyteller: Training-Free Narrative Grounding And Forseebench: Evaluation For Long Form Audio Description, Seung Hyun Hahm

Dartmouth College Master’s Theses

Understanding long-form video requires tracking events, motivations, and relationships across time rather than describing isolated frames. However, existing video--language models (VLMs) and audio description (AD) systems often generate short-horizon descriptions that omit narrative context, causal intent, and story continuity, limiting accessibility for blind and low-vision (BLV) audiences. This thesis investigates how long-form AD can be grounded in narrative memory without relying on expensive supervised training pipelines or heavily curated annotations.

We propose StoryTeller, a training-free retrieval-augmented framework for long-form audio description. Instead of depending solely on frame-level perception, StoryTeller summarizes observations into structured narrative facts that capture who did what …

Dynamic Trust Calibration, Bruno Miranda Henrique

Dartmouth College Ph.D Dissertations

Trust calibration between humans and Artificial Intelligence (AI) is crucial for optimal decision-making in collaborative settings. Excessive trust can lead users to accept AI-generated outputs without question, overlooking critical flaws, while insufficient trust may result in disregarding valuable insights from AI systems, hindering performance. Despite its importance, there is currently no definitive and objective method for measuring trust calibration between humans and AI. Current approaches lack standardization and consistent metrics that can be broadly applied across various contexts, and they don’t distinguish between the formation of opinions and subsequent human decisions. This thesis brings a novel and objective method for …

From Attention To Reasoning: Beyond Accuracy In Multimodal Ai, Wayner Barrios

Dartmouth College Ph.D Dissertations

Multimodal large language models have achieved impressive performance on vision-language benchmarks by integrating visual encoders with large language models. Yet a critical gap persists between benchmark accuracy and genuine multimodal understanding: current evaluation frameworks assess performance by final answers alone, rewarding confident predictions while leaving systematic reasoning failures undetected.

This thesis addresses this gap through a unified framework that progresses from understanding to reasoning, using video as the most comprehensive multimodal testbed. Video inherently combines vision, audio, and language with temporal dynamics and massive token redundancy; techniques developed for video's comprehensive challenges transfer naturally to simpler multimodal tasks.

On understanding …

Ai Interpretability In Healthcare Communication, Ananya Jeyappragash

Dartmouth College Master’s Theses

Artificial intelligence has increasingly been adopted in healthcare, largely for specialized tasks and under significant human oversight. The use of large black-box systems raises important concerns about transparency in high-stakes environments such as clinical decision-making. Clinical communication is fundamentally human-centered, and failures in judgment can have serious consequences for patient care. Overestimating the reasoning abilities of large language models may lead to undue trust in fabricated or “hallucinated” outputs, while rejecting AI-assisted tools altogether may preserve inefficient workflows and contribute to missed or delayed diagnoses. These concerns reflect a broader tradeoff between accuracy and interpretability: although more complex models may …

Designing Narrative-Based Ai Assistance For Sensemaking In Collaborative Environments: Case Studies In Education And Dementia Care, Dylan Edward Moore

Dartmouth College Ph.D Dissertations

This thesis addresses a gap in the human-computer interaction literature regarding the design, development, and evaluation of narrative-based AI assistance for collaborative, complex problem solving. I explore this design space through three case studies across the domains of education and dementia care. This work encompasses multi-year industry partnerships and longitudinal fieldwork, user-centered design, dataset curation, model training, and system evaluation.

Specifically, the first case study considers a story-based web platform for teaching AI literacy through peer-generated, personalized narrative scaffolding. Learners on the platform showed significant knowledge gains and other learning-related outcomes. To describe the novel design of this system, I …

Unsafe2safe: Controllable Image Anonymization For Downstream Utility, Minh Dinh

Dartmouth College Master’s Theses

Large-scale image datasets frequently contain identifiable or sensitive content, raising privacy risks when training models that may memorize and leak such information. We present Unsafe2Safe, a fully automated pipeline that detects privacy-prone images and rewrites only their sensitive regions using multimodally guided diffusion editing. Unsafe2Safe operates in two stages. Stage 1 uses a vision--language model to (i) inspect images for privacy risks, (ii) generate paired private and public captions that respectively include and omit sensitive attributes, and (iii) prompt a large language model to produce structured, identity-neutral edit instructions conditioned on the public caption. Stage 2 employs instruction-driven diffusion editors …

Lens: Llm-Enabled Narrative Synthesis For Mental Health By Aligning Multimodal Sensing With Language Models, Wenxuan Xu

Dartmouth College Master’s Theses

Multimodal health sensing offers rich behavioral signals for assessing mental health, yet translating these numerical time-series measurements into natural language remains challenging. Current LLMs cannot natively ingest long-duration sensor streams, and paired sensor–text datasets are scarce. To address these challenges, we introduce LENS, a framework that aligns multimodal sensing data with language models to generate clinically grounded mental-health narratives. LENS first constructs a large-scale dataset by transforming Ecological Momentary Assessment (EMA) responses related to depression and anxiety symptoms into natural-language descriptions, yielding over 100,000 sensor–text QA pairs from 258 participants. To enable native time-series integration, we train a patch-level encoder …

Anatomy Of An Ai Arms Race: U.S. And China Technological Dispute For Ai Leadership, Denisse I. Rojas Maldonado

Dartmouth College Master’s Theses

The history of societies and the emergence of powerful states have been marked by cycles of conflict and war, followed by periods of cooperation that foster international stability. Similarly, the Cold War era saw a significant rise in military and economic capabilities, which highlighted a security dilemma as the former USSR and the United States sought to protect their national interests. Currently, artificial intelligence has expanded the scope of invisible warfare beyond the atomic bomb. Some scholars like Paul Scharre advocate that there is no arms race in place, and others support the idea of a healthy competition and collaboration …

Revitalization Of Endangered Languages With Ai, Ivory Yang

Dartmouth College Master’s Theses

The preservation and revitalization of endangered languages, particularly those with minimal digital presence, presents significant challenges for computational linguistics. This thesis addresses these challenges by proposing novel methods for language identification and data generation, focusing on underrepresented Indigenous languages, specifically Nüshu, Native American and Native Alaskan languages.

In the first study, a COLING 2025 paper, we present NüshuRescue, an AI-driven framework designed to facilitate the preservation of Nüshu, an endangered script used exclusively by Yao women in China. Using minimal seed data, we demonstrate how GPT-4-Turbo can generate new translations, expanding a publicly available Nüshu-Chinese corpus, achieving 48.69% accuracy in …

Property Testing Ai: An Efficient Frontier, Paul Sopher Lintilhac

Dartmouth College Ph.D Dissertations

In this dissertation, we take a step towards addressing the major problem of a lack of standardized and rigorous approaches to testing and evaluation of AI systems. Taking inspiration from both the fields of Property Testing and Property Based Testing (for programs), we develop a novel taxonomy of partially overlapping classes of properties of AI systems, including simple properties, compound properties, higher order properties, data relation properties, and architecture-utility properties. We argue that this taxonomy categorizes a diverse set of AI traits -- including accuracy, fairness, robustness, monotonicity, point-wise and global privacy properties, sensitivity, and more -- according to the …

Exploring The Facets Of Responsible Ai: Interpretability, Biases, And Morality Of Large Language Models, Sean Xie

Dartmouth College Ph.D Dissertations

This thesis investigates critical aspects of responsible artificial intelligence (AI) — specifically model interpretability, bias detection and mitigation, and moral alignment in large language models (LLMs) — due to their pivotal role in the deployment of transparent, fair, and ethical AI systems. By addressing these dimensions of responsible AI, we hope to foster the increased trust and understanding necessary for wider AI adoption.

We begin by surveying the existing landscape of interpretability metrics and critically assess the effectiveness of interpretability methods designed to generate reliable explanations. Building upon this evaluation, we introduce novel model architectures and frameworks explicitly developed to …

Computational Modeling And Structural Generation Of Piano Music In The Classical Style, Yijing Feng

Dartmouth College Ph.D Dissertations

Listening to fast-tempo piano sonatas of the Classical period (circa 1750-1820) has been shown to have therapeutic effects for neurological disorders such as epilepsy. The limited existing repertoire of music in this style motivates the creation of more long-form, coherent compositions with clearly defined structure. Despite the long history of computer-based music generation and recent progress in deep learning, particularly transformer-based models, generating structurally coherent long-form music remains a major challenge. This difficulty stems from the scarcity of reliable structural annotation datasets, the computational demands of modeling very long musical sequences, and the lack of effective structural encoding in both …

Soft Modular Robots: From Modular Tensegrity Structures To Bioinspired Sea Robots, Luyang Zhao

Dartmouth College Ph.D Dissertations

The rapid advancement of robotics necessitates systems capable of adapting to complex, unstructured environments. Soft robots, with their flexibility and compliance, excel in delicate interactions, making them ideal for medical applications and search-and-rescue missions. Modular robots, on the other hand, offer reconfigurability, enabling diverse task-specific adaptations in dynamic settings. Despite their individual advantages, the integration of soft and modular robotics remains underexplored. This proposal aims to develop soft modular robots that combine the adaptability of soft robotics with the versatility of modularity. These systems will be capable of autonomously transitioning between locomotion, manipulation, and infrastructure assembly across land, water, and …

Examining Intersectional Queer Biases In Large Language Models: A Combined Statistical And Visual-Qualitative Approach For Quantification And Explanation, Huu Duong (Chip) Nguyen

Computer Science Senior Theses

Despite significant advancements in research on (intersectional) social biases in Large Language Models (LLMs), intersectional biases affecting subgroups within the LGBTQ+ community remain critically understudied. Existing bias detection methodologies often prioritize quantification but lack depth in explaining the specific stereotypes/biases that shape evaluation metrics. To address these gaps, this study proposes a combined statistical and visual-qualitative approach to quantify and identify persistent intersectional queer biases in five recent, state-of-the-art LLMs through a downstream story generation task. Findings from analysis uncover substantial evidence of stereotypes that perpetuate harmful, reductive narratives against intersectionally marginalized groups within the LGBTQ+ community. To promote public …

Heartdj - Music Recommendation And Generation Through Biofeedback From Heart Rate Variability, Egemen Şahin

Dartmouth College Master’s Theses

This study investigates the integration of real-time physiological data with AI-generated music to enhance emotional well-being, stress regulation, and focus, using Heart Rate Variability (HRV) as a biomarker of autonomic function. Conducted in two phases—Stable Audio Open (SAO) and Suno (SUNO)—the research evaluates biofeedback-driven music interventions across varying daily music-listening habits.

In the SAO phase, short AI-generated instrumental tracks were compared with Spotify recommendations and guided meditation. Modest HRV improvements were observed in biofeedback conditions, but participants noted emotional limitations, citing short track lengths and abrupt transitions.

The SUNO phase addressed these limitations with longer, more complex AI-generated compositions combined …

Incorporating Visual Information Into Natural Language Processing, Maxwell Mbabilla Aladago

Dartmouth College Ph.D Dissertations

Natural language describes entities in the world, some real and some abstract. It is also common practice to complement human learning of natural language with visual cues. This is evident in the heavily graphical nature of children’s literature which underscores the importance of visual cues in language acquisition. Similarly, the notion of “visual learners” is well recognized, reflecting the understanding that visual signals such as illustrations, gestures, and depictions effectively supplement language. In machine learning, two primary paradigms have emerged for training systems involving natural language. The first paradigm encompasses setups where pre-training and downstream tasks are exclusively in natural …

Openmuse: Integrating Open-Source Models Into Music Creation Workflows, Tyler K. Vergho

Dartmouth College Master’s Theses

This master's thesis introduces OpenMUSE (Open Multimodal Unified Sound Engine), a platform that demonstrates the potential of open-source AI music generation by integrating state-of-the-art deep learning models into a unified system. By unifying ten different open-source models, including MusicGen, AudioLDM2, and custom-trained text-to-symbolic music generation models, OpenMUSE aims to create a user-friendly interface that empowers artists to produce complex, adaptive musical compositions. The system enhances accessibility by providing a simple web interface and natural language controls, while improving controllability through features like melody conditioning and semantic audio editing. Specifically, OpenMUSE offers a digital audio workstation (DAW)-inspired interface that lowers the …

Motion Planning For A Flexible Modular Raft Robot, Chun-Yi She

Dartmouth College Master’s Theses

This thesis presents a hierarchical motion planning framework for SoftRafts, a modular and deformable aquatic robot capable of performing locomotion and manipulation tasks on water surfaces. SoftRafts consist of soft and rigid components that enable structural reconfiguration, offering adaptability in unstructured aquatic environments.

To address the complexity of planning in high-dimensional, deformable systems, the proposed method uses a bounding-shape abstraction, specifically, enclosing circles and rectangular bounding boxes to simplify motion planning. These enclosures abstract the robot's overall shape, reducing the high-dimensional planning problem into a lower-dimensional problem. A global planner uses a probabilistic roadmap (PRM) to compute a collision-free path …

Robust And Generalizable Representations In The Hippocampus, Hung-Tu Chen

Dartmouth College Ph.D Dissertations

An intelligent system must balance generalizing across similar experiences with maintaining the distinctiveness of each experience. This thesis explores how the hippocampus manages this balance through its neural representations to support adaptive behavior. In Chapter 1, I provide an overview of key hippocampal phenomena that contribute to this process, including remapping, splitter signal, and replay. In Chapter 2, I challenge the concept of random remapping by showing that it is possible to predict, better than chance, how a given experience will be encoded in the hippocampus across different subjects. This suggests that encoding of related experiences, which was previously thought …

Achieving Domain-Independent Certified Robustness Via Knowledge Continuity, Alan Wenyuan Sun

Computer Science Senior Theses

We present knowledge continuity, a novel definition inspired by Lipschitz continuity which aims to certify the robustness of neural networks across input domains (such as continuous and discrete domains in vision and language, respectively). Most existing approaches that seek to certify robustness, especially Lipschitz continuity, lie within the continuous domain with norm and distribution-dependent guarantees. In contrast, our proposed definition yields certification guarantees that depend only on the loss function and the intermediate learned metric spaces of the neural network. These bounds are independent of domain modality, norms, and distribution. We further demonstrate that the expressiveness of a model …

Impact Of Similarities In Gender And Physical Appearance Between User And Embodied Conversational Agents On Trustworthiness, Empathy, And Service Evaluation, Sookyoung Park

Dartmouth College Master’s Theses

Embodied conversational agents (ECAs) have significantly enhanced human-machine interactions and show considerable potential in various industries such as customer service, education, healthcare, entertainment, and finance [1, 2]. This study explores the impact of similarities in gender and physical appearance between ECAs and users on the perceptions of trustworthiness, empathy, and service evaluation within the context of counselor ECAs. We conducted a within-subject experiment (n=50), using a 2x2 factorial arrangement, that varied the gender and the physical appearance of four distinct AI avatars. Participants interacted with each avatar, completing a post-experiment survey and participating in semi-structured interviews. Our findings indicate that …

Towards Machine Proficiency With Semantic Underspecification, Zachary S. Gottesman

Dartmouth College Master’s Theses

Human natural language communication frequently relies on extra-linguistic information to fill in gaps in the linguistic signal left by semantic underspecification, or the omission of details that can be inferred from prior knowledge or other modalities. Underspecification is particularly common in conversations between acquaintances, since these interlocutors share context. Underspecification is a key and beneficial feature of natural language that improves efficiency, although it can cause communication to fail if it is not resolved correctly. For language models to communicate effectively and in a human-like fashion, they must learn how to recognize and utilize underspecified language. This thesis argues that …

Toward The Integration Of Behavioral Sensing And Artificial Intelligence, Subigya K. Nepal

Dartmouth College Ph.D Dissertations

The integration of behavioral sensing and Artificial Intelligence (AI) has increasingly proven invaluable across various domains, offering profound insights into human behavior, enhancing mental health monitoring, and optimizing workplace productivity. This thesis presents five pivotal studies that employ smartphone, wearable, and laptop-based sensing to explore and push the boundaries of what these technologies can achieve in real-world settings. This body of work explores the innovative and practical applications of AI and behavioral sensing to capture and analyze data for diverse purposes. The first part of the thesis comprises longitudinal studies on behavioral sensing, providing a detailed, long-term view of how …

Towards Scalable Autonomous Underwater Construction With Free-Floating Robots, Samuel Eric Lensgraf

Dartmouth College Ph.D Dissertations

This thesis presents the first free-floating autonomous underwater construction system. Our system built structures weighing up to 100Kg (75Kg in water). Our robot builds structures made of standard cinder blocks and custom designed interlocking cement blocks. It is the first construction robot that uses active buoyancy compensation to efficiently transport building materials. It is also the first construction robot that can reconfigure visual fiducial markers on a foundation during the construction process to expand its working area.

Underwater construction is a challenging problem for free-floating robots. Currents can buffet the robot, and visibility conditions can change. We focus on achieving …

Automated Cinematographer For Vr Viewing Experiences, Zihan Wu

Dartmouth College Master’s Theses

As the virtual reality (VR) industry continues to evolve, the question of how to effectively capture VR experiences for an audience remains a challenge. The predominant method of showcasing VR applications through first-person recordings lacks cinematic interest, failing to capture other viewpoints and the essence of the moment. Meanwhile, manually setting up cameras and editing videos requires technical expertise on behalf of the user. In this paper, we propose the use of machine learning (ML) to automatically select the most compelling predefined viewpoint in a VR environment, at any given moment. Our models, trained on actor motion and voice volume, …

3-D Reconstruction For Underwater Robots With A Monocular Camera And Lights, Monika Roznere

Dartmouth College Ph.D Dissertations

Before a robot can act, it must perceive its environment. Though, this is not a simple task when considering the challenges in underwater domains -- poor visibility conditions, limited sensor configurations, and lack of readily accessible localization. Underwater robots have, nevertheless, improved dramatically with more extensive sensor and navigation equipment. Robot and sensor use have enabled us to explore all reaches of our oceans. On the other hand, these same robots are not easily accessible or transferable to many practical tasks, including fishery management, infrastructure maintenance, disaster response, site conservation, and ecological surveys. There is a growing need for robots …

Design, Analysis, And Drop Assembly Of Interlocking Rigid Bodies, Amy K. Sniffen

Dartmouth College Ph.D Dissertations

This work presents a system of interlocking blocks that can be used to build a wide variety of structures. The blocks slide together to form structures that interlock geometrically like a puzzle to form semi-permanent structures without the need for cement or friction lock. The blocks are designed to be easy to fabricate, assemble, and disassemble. Contributions of the block designs include a novel interlocking joint structure; the joints are wedge-shaped, allowing for error mitigation during assembly and allowing structures to be assembled without jamming even if there is manufacturing error. We introduce planar, 3D, and volumetric designs using these …