The recognition of individuals as sufficiently "Darwinian" entities for natural selection has led to the exploration of clades as alternative units of selection to species in the field of philosophy of biology. Previous research has demonstrated that clades meet the criteria of "Darwinian" individuals, intriguing philosophers to explain natural selection at the clade level. This research employs an Abstract Data Type (ADT) approach from computer science, constructing a data structure (DS), and attempts to implement the clades ADT both with and without species. The objective is to scrutinize the inner structure of clade and unearth the interplay between clades and species, a relationship not explicitly implied in their definitions. Preliminary findings affirm the technical feasibility of implementing clades ADT as a clade-tree data structure with species, while the feature of each level of clades remains unclear: (1) Issues of trait identification emerge when assigning properties to organisms within clades; (2) The empirical nature of the data used to construct the data structure representing clades limits holistic demonstration of evolution. On the other hand, a clade-tree data structure without species encounters challenges: (1) Historical organisms within inner stages cannot be distinctly separated without defined species-like boundaries (2) If properties are specified to species-level, the species-free intention of the data structure becomes questionable. The emergence of a dependency on species introduces a compelling reason for reconsidering the ability of clades in explaining natural selection, particularly for those philosophers who oppose species as a unit of selection.

There is currently a lack of tractable data on what was published in England in the early 20th century. However, this information exists in printed volumes of The English Catalogue of Books, which have been digitized through the HathiTrust digital library. The English Catalogue of Books, released in the UK by the trade publication Publishers’ Circular, provides a yearly record of books published from the mid-19th to the early 20th century. Each catalog has been converted to plain text automatically through optical character recognition (OCR). Our aim was to parse this plain text into user-friendly data on books published each year in the UK. However, the OCR-generated text often contained errors and inconsistencies that prevented the effective extraction of data on books listed in the catalogs. Thus, we aimed to gauge the accuracy of existing methods for parsing the catalogs and to tailor processes for gathering data from catalogs published between 1908 to 1922. By writing regular expressions to capture, split, and match the patterns of bibliographic entries, we were able to improve the accuracy of processes for extracting data. Our solutions increased the number of publications for which information was accurately being captured by on average 28%, and we were able to record information on over 21,500 books that previously had not been captured. We also created a summary dataset from the catalogs with information on the overall output of the publishing industry during these years. By analyzing and visualizing this publishing data, we were able to show that fiction was the most frequently published genre during the period. We anticipate this project to be the foundation of more work towards efficiently parsing The English Catalogue of Books in order to offer insights into the British publishing industry in the 19th and 20th centuries.

Addiction is characterized by the compulsive use of substances despite adverse consequences, a process closely linked to dopamine-induced changes in the Nucleus Accumbens (NAc) and its role as the brain's "reward center." The NAc integrates information from various brain regions, including the Paraventricular Thalamus (PVT), to produce motivated behaviors. Recent studies have identified the PVT, especially its anterior segment (aPVT), as a critical hub in addiction neurocircuitry, but findings have been inconsistent, likely due to the PVT's heterogeneity and the specific neurochemical and anatomical properties of its connections to the NAc. Prior research has shown that aPVT neurons, identifiable by neurotensin expression, send excitatory projections to the NAc, which are modulated by endogenous cannabinoids (eCBs). These interactions suggest a complex regulatory mechanism. Preliminary experiments used techniques including transsynaptic viral tracing and in vivo calcium imaging, to study the activity dynamics of NAc neurons, particularly those expressing Proenkephalin (PENK) and receiving aPVT inputs, during reward-seeking tasks. I propose to extend these findings by employing a multidisciplinary approach that combines experimental neuroscience with sophisticated computational analysis. By applying dimensionality reduction techniques, clustering algorithms, and machine learning models to neural and behavioral data, I aim to map the functional connectivity within the NAc and elucidate the roles of specific neuronal ensembles in reward-seeking behavior. This comprehensive analysis will not only clarify the neurobiological underpinnings of addiction but also contribute to the development of targeted therapies for addiction and related disorders, leveraging the unique intersection of computational neuroscience and behavioral analysis.

This project aims to document the phonology (sound systems) of Triestin, a Venetian dialect spoken by about 200,000 people in Trieste, Italy. Triestin lies at the intersection of the Germanic, Romance, and Slavic language families and thus exhibits many unique phonological phenomena. The documentation of these phenomena is increasingly important as the number of native speakers rapidly decreases. I recorded four native Triestin speakers reading over 300 words each to elicit a variety of words and sentences. In my illustration, I analyze two main phenomena: (i) the vowel system and (ii) intonation. To analyze the vowel system of Triestin, I took 100 words with vowels in different phonological environments (surrounding sounds) and analyzed patterns in their formant values (resonant frequencies that distinguish one vowel from another). I found the vowel system is unique in its reduced size, having only 5 distinct vowels as opposed to other Venetian dialects’ 7 vowels. My analysis describes how this reduced vowel system exhibits variations unique to the dialect, such as raising in stressed penultimate syllables. Triestin sentences also have a rich system of intonation that differs significantly from other Venetian dialects. In my research, I demonstrate how declarative sentences’ intonation is a function of the pragmatics (contextual meaning) of the conversation. The data also suggests that the intonation interacts with the stress of the final word in the sentences, resulting in greater rising intonation for sentences ending in a stressed syllable. The unofficial languages of Italy, including Venetian and its dialects, are extremely under-documented. Projects such as my illustration of Triestin phonology help with expanding the body of literature on the unique features of endangered dialects. This project is also the first step in a more holistic documentation of Triestin, with future projects aimed at studying the syntax and sociolinguistics of the language.

Across the world, there are roughly 70 million Deaf and hard of hearing (DHH) people who use over 200 different sign languages. The ability to look up the meaning of unknown words is important for any language, and there is currently no easy way to look up signs due to sign languages having no common written form–meaning you can’t just type a sign into your keyboard. One common approach to sign language search is feature-based lookup, which allows users to search for a sign by inputting its visual features: handshape, location, palm orientation, and movement. Previous approaches to feature-based lookup have several limitations in that they often have poor matching of features to signs, don’t allow for the user to omit the features they don’t know, and have cumbersome search interfaces. The work done in Bragg et al. (2015) took a step towards addressing these issues by introducing ASL-Search—a robust search system that compares a user’s sign query to a database of labeled queries inputted by previous users using a topic modeling method called Latent Semantic Analysis (LSA) and cosine similarity—but their system still requires the user to choose between a large set of 178 features. In part 1 of our study, we found that by dropping certain categories of features, and combining/reducing others into similar categories, we could dramatically reduce the number of features needed to search for signs from 178 to around 45 to 65 without significantly impacting the accuracy the results returned during search. In reducing the number of features needed, we came up with smaller, intuitive sets of handshapes and movements. In part 2 of our study, we are evaluating how usable our new handshape and movement features are. By making it possible to search for the meaning of signs using less features, we hope to make sign language search interfaces usable enough to be used in the real world.

Artificial intelligence has made significant progress in image classification, an essential task for machine perception to achieve high-level image understanding. Despite recent advances in vision-language fields, image classification is still challenging because classes are inherently multimodal. Traditional approaches provide solely images or descriptions of a given class, but this does not properly encapsulate the entire concept. I aim to understand if utilizing multiple modalities when classifying images will allow for more robust and generalizable image classification. Here, I leverage Ensemble Learning to form a weighted sum of the embeddings produced by my classifiers. The ensemble consists of an image-to-text classifier that utilizes generated class descriptions and an image-to-image classifier that uses generated images for prediction. My results show that this improved conceptual representation from the multimodal approach directly results in higher classification accuracy. Additionally, this approach allows for these improvements without any additional context, training, or fine-tuning, meaning this method is completely zero-shot. Improved image classification directly impacts a myriad of fields, and enhances how humans use computing technology. Whether it be more accurate diagnoses in medical imaging or robust crop monitoring in agriculture, improvements in image classification extend across disciplines and aid the quality of life for all.

The measurement problem is the challenge of reconciling the probabilistic and deterministic aspects of Quantum Mechanics. In this study, our primary aim is to unravel the measurement problem by investigating the possibility that quantum collapse occurs according to the probabilities presented by Born's Rule due to a perturbation on the system, appearing during the measurement process. Our group employs a multifaceted approach, where we examine the interplay between time dynamics and classical limits, alongside the influence of time-(in)dependent perturbations. Computational simulations serve as our primary tool in this exploration. One part of our group worked with a 3-well system with a time-independent perturbation, another part looked at a 2-well system with a time-dependent perturbation, and the last part saw what a time-independent perturbation does to a 2D-well system. We anticipate that our investigations will uncover critical parameters that are expected to yield probabilities consistent with Born's Rule, a foundational principle of Quantum Mechancis. This research points towards a potential reconsideration of quantum collapse as a dynamical, affected by the perturbation introduced by measurement. While preliminary, these findings contribute to the ongoing discourse in Quantum Mechanics and may offer insights for future theoretical developments and applications.

Data visualizations are critical to understanding large, complex datasets. For example, visualizations help us detect missing and erroneous data values, identify potential relationships between data variables, and review the output of machine learning models. That being said, customized visualizations can be difficult to create, because they require the use of specialized toolkits such as ggplot2, Vega-Lite, or D3. In this research, we study how people use a browser-based toolkit called D3 to program custom visualizations, with the long-term goal of creating AI-assistants to help people code in D3. However, to train rigorous AI models, we need a large input corpus of D3 examples that is accurate and reliable. In this presentation, we share our progress towards building this training corpus. First, we mined hundreds of real-world examples from the web. Then, we analyzed these examples to understand how visualization users take complex D3 programs and break them down into easily understandable parts, which we call components. Currently, we are investigating how these different components can be remixed and reused to create D3 visualizations for datasets that a model may not have encountered before. Future research will involve collecting more examples to grow our corpus, and training AI to use the corpus to generate documentation and relevant examples to help new visualization users better understand D3.

In response to the growing importance of inclusive design and accessibility in the realm of software development, this project integrates accessibility concepts and design thinking into a college-level data structures and algorithms course. Our goal is for students to explore accessible design practices, evaluate the impacts of their software design work using accessibility frameworks, and gain an understanding of the social and historical context of disability with technology. Moreover, we aim to motivate how these accessibility concerns are not just an addendum or afterthought, but core and rigorous considerations when building software. We plan to enhance the learning experience and outcomes of students by developing accessibility-focused design activities that help teach design principles and iterative design processes, as well as discussion questions framed around the evaluation and critique of assistive technologies. Through the integration and evaluation of accessibility concepts throughout the curriculum, we expect the enhancement of students' ability to identify design shortfalls through assumptions and functionalities of interfaces and propose alternative designs that consider diverse user needs. Anticipated outcomes include improved student understanding of accessible design practices, increased awareness of the impact of design decisions on user experience using real-world scenarios, and enhanced skills in evaluating software designs using accessibility frameworks. By equipping students with the ability to understand and address the diverse needs of users, this approach not only prepares students for the complexities of real-world challenges but also empowers them to critically evaluate and refine designs in professional settings. Ultimately, it enables the creation of technology that is inherently more supportive and inclusive for all individuals. With this project, we plan to further refine and evaluate the effectiveness of the integrated curriculum from educator and student feedback, as well as make the materials usable for other educators through platforms such as the Teach Access repository.

In the rapidly evolving landscape of remote work, the challenges associated with recalling important information from meetings and missing meetings have increased. One potential solution is to use large language models (LLMs) to summarize meetings to help participants catch up after meetings are over. To have a better understanding of this topic, we systematically reviewed 17 existing commercial tools and research prototypes for LLM-generated meeting summaries. The results show that existing solutions fell short of supporting users to verify and validate the comprehensiveness and accuracy of the generated summary, hindering users from trusting the summary. To address this, the project aims to design and build a more trustworthy LLM-generated meeting summary tool. Specifically, we propose that LLM-generated summary should progressively display relevant meeting information based on the importance of the information and the user’s goals, and include trustworthiness cues to aid users in making accurate trust judgments of the summary. Our preliminary interviews and a literature review showed that users are more hesitant to trust the AI summary when the information is consequential, such as when they missed the meeting or specific action items. While trustworthiness cues such as quotes or links to raw transcripts could increase users’ trust, irrelevant and redundant information erodes people’s trust. To further validate these observations, we will conduct a formative interview study. We will show participants mid-fidelity prototypes exemplifying the key design decisions and elicit their feedback on appropriate trustworthiness cues, desired ways to indicate their goals and intentions, and expectations on the importance of different portions of a summary. These empirically supported insights will inform the final design of a trustworthy LLM-generated meeting summary tool, which we plan to implement and evaluate in the next step.

General anesthesia (GA) is administered as a sedative in nearly 60,000 surgeries daily in the United States. Yet, there is a very limited understanding about how GA impacts brain activity, leading to induced loss of consciousness and pain sensation. Preliminary work in the Heshmati lab has highlighted key subcortical structures that are engaged during anesthesia, but it remains unclear how activity in these regions and across the brain regulates awareness or pain sensation as anesthesia is induced (“induction”), maintained at a steady state (“maintenance”) and removed (“emergence”), as is done during surgeries. My work aims to identify the neural circuits that regulate the loss of consciousness and pain sensation during GA by recording local field potentials (LFP) from mice as they undergo volatile anesthetic isoflurane (ISO). During LFP recordings, I will insert small electrodes into highlighted regions of interest, to capture low-frequency extracellular voltage signals generated by the synchronized activity of nearby neural populations during the three periods of interest: induction, maintenance, and emergence from isoflurane GA. I will analyze the amplitude fluctuations and frequency patterns to identify synchronized oscillations within subregions and assess the level of synchrony, or coherence, across different regions. Given previous findings on the shared and opposed involvement of subcortical regions in pain and anesthesia, I expect to observe coherence among some of the regions, such as the amygdala and hypothalamus, but potentially anti-correlation within specific subsections, such as central vs. basolateral amygdala. Through these experiments, I will be able to monitor the effects of isoflurane anesthesia through a temporally-defined electrophysiological lens, capturing real-time activation dynamics of large neural populations across induction and recovery from anesthesia. Thus, my research aims to further develop our understanding of the brain under GA, by providing novel insight into the neural circuits regulating wakefulness and pain during surgical procedures.

In this research project, we delved into the realm of gate-based quantum computation with a focus on qudit-based quantum computation. In the era of Noisy Intermediate-Scale Quantum (NISQ) computation, there are many avenues for physical implementations of qudits, such as trapped ions, superconducting circuits, and photonic systems. We primarily studied trapped ion qudit-based computation, investigating the notion of universality and how arbitrary gate operations can be simulated by experimentally realizable transformations in such systems. More quantitatively, we analyzed the fidelity under the assumptions of rotation angle errors in trapped ion implementations of quantum gates. We proved several lower bounds for various connectivity graph designs applicable to the 5-level calcium ion under this model of assumptions. Our techniques also generalize to physical systems with more than 5 levels. Currently, our attention is directed toward understanding the impact of entanglement on the aforementioned dynamics and studying the notion of universality for multi-qudit systems. A related question we are trying to answer is how qubit circuits can be converted into qudit circuits to reduce a well-defined notion of "circuit complexity".

A critical feature of autonomous cars is the ability to follow a road or predefined path. Classical methods often rely on extensive prior mapping with precise GPS positioning. These methods are labor intensive and struggle with changing, unstructured environments. Instead, machine learning (ML) models are trained to recognize paths and follow directions. In this work, we combine simulated and real-world data to train a neural network policy that controls an autonomous ground vehicle down a hallway, avoiding collisions. Training a ML road-following model consists of three steps: data collection and preprocessing, model training, and model evaluation. While all three steps pose challenges, collecting high-quality, real-world data can be expensive and dangerous in road environments. Because of this, simulator data is useful as it allows for data to be collected safely and inexpensively. Thus, we study how much the required amount of real-world data can be reduced to successfully train a road-following robot with the use of simulator data. So, we collected simulator data using AirSim to train a convolutional neural network that follows a path in simulation through live environment images. We then fine-tuned the model using real-world data collected from MuSHR cars through hallways of a building. Next, we test the fine-tuned model on the simulator to ensure limited degradation to the model solely trained from AirSim data. Finally, we deploy the model on a robotic car in a real-world environment and evaluate the model’s performance compared to the baseline model trained on real-world data. We demonstrate that we can successfully train a model in simulation (MSE <= 0.01radians), and we expect to show a comparable performance in reducing the number of collisions and minimizing trajectory differences between expert and learned controller from a model trained on simulator + less real-world data and a model trained solely on real-world data.

Multiple Particle Tracking (MPT) is a powerful technique for studying the behavior of microscopic particles, such as viruses and nanoparticles, by tracking individual displacement and movement. One application of MPT is to measure microstructural changes in the brain extracellular environment (ECM) in development and aging, and in response to disease onset and progression. MPT of nanoparticle probes results in the generation of thousands of individual nanoparticle trajectories, from which geometric features, diffusion coefficients, and viscosities can be extracted. The vast array of trajectories contained within our dataset presents a good opportunity for integration into deep learning models that contains self-supervised learning, equivariant graph neural network, and Equivariant transformer. However, to enable MPT data to be trainable and predictable by deep learning models, we need to curate the data to be readable and useable by these models. To enable this, I have created a database and developed a data architecture that would allow MPT data to be passed into Deep learning models that use various techniques such as transformers. I am currently working on utilizing the data architecture on a Deep Learning model that uses transformers and self-supervised learning to predict trajectories of MPT particles. From this model, my expected accuracy of prediction of the trajectories for the MPT data is around 85%. This can allow us to learn complex features directly from raw MPT trajectory data, improve our predictions, and extract biological insights. The python package with our data architecture, the various SQL scripts, and the model will be provided as an open-source resource, allowing other researchers to expand upon my code and apply their unique modifications based on their own data and trajectories.

One of the core problems in the analysis of protein tandem mass spectrometry data is the peptide assignment problem: determining, for each observed spectrum, the peptide sequence that was responsible for generating the spectrum. Two primary classes of methods are used to solve this problem:database search and de novo peptide sequencing. State-of-the-art methods for de novo sequencing employ machine learning methods, whereas most database search engines use hand-designed score functions to evaluate the quality of a match between an observed spectrum and a candidate peptide from the database. We hypothesize that machine learning models for de novo sequencing implicitly learn a score function that captures the relationship between peptides and spectra, and thus may be re-purposed as a score function for database search. Because this score function is trained from massive amounts of mass spectrometry data, it could potentially outperform existing, hand-designed database search tools. To test this hypothesis, we re-engineered Casanovo, which has been shown to provide state-of-the-art de novo sequencing capabilities, to assign scores to given peptide-spectrum pairs. We then evaluated the statistical power of this Casanovo score function, Casanovo-DB, to detect peptides on a benchmark of three mass spectrometry runs from three different species. Our results show that, at a 1% peptide-level false discovery rate threshold, Casanovo-DB outperforms existing hand-designed score functions by 35% to 88%. In addition, we show that re-scoring with the Percolator post-processor benefits Casanovo-DB more than other score functions, further increasing the number of detected peptides.

Large language models (LMs) are prone to generate diverse factually incorrect statements, which are widely called hallucinations. Current approaches predominantly focus on coarse-grained automatic hallucination detection or editing, overlooking nuanced error levels. In this project, we propose a novel task—automatic fine-grained hallucination detection—and present a comprehensive taxonomy encompassing six hierarchically defined types of hallucination. To facilitate evaluation, we introduce a new benchmark that includes fine-grained human judgments on two LM outputs across various domains. To run this evaluation, I directly managed the collection of around 400 total human annotations which were analyzed to better understand the hallucinations present in LM outputs. My analysis using this benchmark reveals that ChatGPT and Llama2-Chat exhibit hallucinations in 60% and 75% of their outputs, respectively. A majority of these hallucinations fall into categories that have been underexplored in previous work. As an initial step to address this, I trained FAVA, a retrieval-augmented LM by carefully designing synthetic data generations to detect and correct fine-grained hallucinations. I set up the synthetic data generation pipeline to train FAVA which consists of prompting ChatGPT to noise a passage and insert errors one by one. The noisy passage is then post processed into our training erroneous input and edited output pairs. On our benchmark, our automatic and human evaluations show that FAVA significantly outperforms ChatGPT on fine-grained hallucination detection by a large margin though a large room for future improvement still exists. FAVA’s suggested edits also improve the factuality of LM-generated text, resulting in 5-10% FActScore improvements. These results further demonstrate the strong capabilities of FAVA in detecting factual errors in LM outputs.

Addressing bias in artificial intelligence (AI) and machine learning (ML) systems is crucial for ensuring fairness, transparency, and ethical integrity. This study introduces a pioneering interdisciplinary approach, blending advanced computational methods with social sciences insights to tackle the multifaceted nature of bias. Through a mixed methods strategy that combines quantitative and qualitative data, we scrutinize algorithmic outcomes and conduct different case studies of stakeholders—developers, users, and communities affected by AI/ML biases. Our initial findings indicate that bias transcends technical boundaries, manifesting as a complex socio-technical dilemma that demands both algorithmic adjustments and societal reforms. We highlight specific biases, such as gender and racial disparities in recruitment algorithms and facial recognition technologies, underscoring the critical need for our research. To address these biases, we propose adopting data enhancement techniques, fairness-focused learning algorithms, and promoting explainable AI practices. Inspired by influential figures like Joy Buolamwini, founder of the Algorithmic Justice League, and Cathy O'Neil, author of Weapons of Math Destruction, we emphasize the importance of inclusive datasets and critically examining opaque algorithms. Our future efforts concentrate on developing comprehensive guidelines to reduce AI/ML biases and exploring the broader societal impacts of establishing unbiased AI and ML systems. By cultivating more equitable and ethical AI and ML frameworks, our research aims to meet the diverse needs of global communities, setting a new standard for responsible AI development.

I investigate the influence of cultural and linguistic backgrounds on visual perception and semantic interpretation within computer vision. This study addresses the question: Are there significant variations in the semantic content described by vision-language datasets and models across different languages? Guided by the hypothesis that cultural and linguistic diversities lead to distinct semantic interpretations, I compare multilingual datasets against monolingual counterparts. I developed metrics such as scene graph complexity, embedding space width, and linguistic diversity to quantify semantic variations across languages in both human-annotated and model-generated image captions. The methodology involves using linguistic tools and translation techniques to ensure semantic consistency across languages. Our findings indicate that multilingual captions contain, on average, 21.8% more objects, 24.5% more relations, and 27.1% more attributes than monolingual ones. Furthermore, models trained on diverse linguistic content demonstrate improved generalizability across different linguistic datasets. This study contributes to the understanding of how language and culture impact visual perception in computer vision and advocates for more inclusive dataset compilation and model training strategies.

According to the Kaiser Family Foundation, almost half of the US population resides in areas with a shortage of mental health professionals. The United States Census Bureau states that 26 million Americans do not have health insurance. Without insurance, a single mental health session can cost hundreds of dollars. Due to these factors, people are looking for alternative ways to address their mental health issues. The purpose of the research being conducted is to create conversational AIs that provide many benefits that a psychotherapist could offer, such as engaging the user in deep conversations, building relationships with the user, and providing apt responses, while avoiding some demerits like patient disclosure and preconceived bias. The research being conducted makes use of Machine Learning and Natural Language Processing models in its development, which is unlike most other popular mental health chatbots that uses a decision-based approach to find the best responses. The anticipated goal of our study is to better understand user comfortability and challenges with the AI therapist that is in development and how to better optimize it for a wider audience. By developing an AI solution that provides the benefits of a psychotherapist, we can better address the mental health crisis and worker shortage in America that predominantly affects low-income and underrepresented communities. The expected result would be a live chatbot application that can understand the mental health issues of the user, and provide advice and suggestions relevant to the user's concerns and issues.

Healthcare workers struggle to find a work-sleep balance, especially during periods of night shifts which according to Likitha Shaik (a clinical researcher at Hennepin Healthcare), results in a 37.7% increase in patient harm and medical errors, highlighting the urgent need for effective solutions. Researchers have proposed polyphasic sleep cycles as a solution to optimize healthcare workers’ sleep by managing accumulating sleep debt. Polyphasic sleep consists of two or more sleeping periods a day that happens to align with the chaotic and unpredictable schedules of healthcare workers, especially those in emergency medicine, trauma surgery, neurosurgery, OB/GYN, and critical care medicine. This literature review investigates how biological and lifestyle factors affect sleep optimization in healthcare workers. Biological factors include age, circadian rhythm, melatonin production, and sleep disorders. Lifestyle factors include physical activity, nutrition and hydration, stress and mental health, and temperature regulation. I constructed this literature review through research using peer-reviewed and popular sources that highlighted the connection between said factors and sleep optimization. The findings of this literature review demonstrate how healthcare workers can best incorporate polyphasic sleep cycles in consideration of the aforementioned biological and lifestyle factors. By understanding how individual factors affect the optimal use of polyphasic sleep, we can start incorporating these findings into staff training and medical school curriculums, especially regarding sleep medicine. Consumer sleep technologies, such as sleep-tracking smartwatches, will also greatly benefit from this research as they can better meet the specific needs of an individual and can be used more frequently in clinical settings. Implementation of these findings is not only crucial to decrease medical errors while enhancing alertness and performance quality among healthcare workers, but also to potentially set new standards for workplace well-being across various professions requiring extended night shifts and/or chaotic sleep schedules.

Understanding how DNA folds in three dimensions is crucial for deciphering cellular function. Chromosomal contacts are interactions between different DNA regions. These contacts hold key information about tissue-specific characteristics, such as gene expression and regulation. However, current predictive models for genome folding primarily focus on within-chromosome interactions, largely ignoring variations across tissues and the role of interactions between chromosomes (trans-contacts). To address these issues, we developed TwinC, a machine learning model that predicts trans-contact maps from pairs of nucleotide sequences. To build TwinC, we used a convolutional decoder coupled with an encoder architecture that can be configured to employ transformers, convolutional networks, or a hybrid approach. Preliminary results suggest that the convolutional architecture achieves performance comparable to Orca, the current state-of-the-art in sequence-to-contact predictions. TwinC is trained and evaluated on contacts measured in two human tissues and one mouse tissue. We are experimenting further with other encoder architectures, fine-tuning the model, and investigating how it generates its predictions. This research will provide valuable insights into the underlying biological mechanisms responsible for chromosomal contacts and lead to an improved, high-performance model for predicting trans-contacts.

Titanium dioxide is a compound that has been used in the realm of nanotechnologies for decades. Titanium dioxide is used as a protective and high-refractive index optical coating. It also has strong mechanical and chemical stability. This is also true of other ceramics throughout the Washington Nanofabrication Facility with compounds such as aluminum oxide, titanium nitride, and aluminum nitride. With this in mind, the goal is to explore the best conditions to reactively sputter titanium dioxide. Thus, I developed a series of experiments including a screening for significant factors, and a following set of experiments using previous results to find an optimum point. In order to produce titanium dioxide in a physical vapor deposition environment such as the sputter tool utilized, the chamber holding the designated surface was pumped down and exposed to a chamber filled with argon, whose function was to hit the target titanium and a percentage of oxygen designed to react with the titanium in order to form titanium dioxide. By generating experiments designed to better understand how titanium and oxygen would react within the sputter tool, there was an aim to better screen for factors and understand the surface composition of the titanium dioxide. The desired outcome of this research is to build a working titanium dioxide recipe that optimizes deposition. Based on preliminary testing, the ideal outcome is likely produced under low pressure and higher power conditions. Future goals may include uniformity and accuracy within titanium dioxide, but also with other materials. With the implementation of this recipe, which is both optimized for the deposition of the material as well as its applicability for the lab, other recipes utilizing similar methods are desired. With a working titanium dioxide recipe, titanium and aluminum nitride recipes can better be developed.
 

Evolution is a challenging topic that high school students often struggle to grasp. The yEvo project, a collaboration between genetics labs and high school biology classrooms, seeks to address this issue by giving students an active hands-on experience in evolution using yeast as a model. Over the course of a school year, yEvo students expose yeast to a selective pressure (such as antifungal drugs), and the natural mutation rate creates beneficial mutations that increase in frequency due to selection. Whole-genome sequencing and computational analysis of the ancestor versus evolved populations reveals changes in the genome that were selected over the course of the experiment. However, the resulting raw mutation data is difficult for students to interpret, thus motivating a need for an intuitive, interactive method to visualize sequencing results. Built using R shiny, a first draft of the yEvo mutation browser consisted of a chromosome diagram with each classroom’s mutation data mapped, a pie chart of mutation types, and a gene viewer depicting the altered sites in each mutated gene. Our work focuses on upgrading the user interface to make it more intuitive, optimizing the backend data processing for more streamlined data filtering, and adding features that allow users to upload and interact with their own datasets. This tool will enable students to more easily interpret the results of their evolution projects. Students will be able to view their classroom data, compare it to other datasets with the same condition, visualize mutation clusters in the genome, and see how each mutation changed the genes. In the future, the yEvo browser can be outsourced and used as a framework for data visualization for other model organisms that can serve to benefit the genetics community as well as educators.

In the context of computer-aided design, researchers have studied how to reconstruct an input geometry in CAD by decomposing it into CAD primitives. Such reconstruction is useful for creating CAD designs for manufacturing applications. What we want to study is also object decomposition but towards a different goal: understanding object affordances and interactability. For example, a handle of a basket can be grasped or hung from a sticky hook, and we recognize this affordance or functionality because it has a certain shape (e.g. hook or rod). Prior research has identified eight types of shape primitives that are common in everyday objects, but the existing tagging process requires a high degree of modeling expertise. We aim to create a more automatic and easy-to-use tagging tool. Our proposed research is to develop user-in-the-loop methods for tagging shape primitives given an object geometry. This takes advantage of human intuition for how objects function and interact. We start with building an interface, where users sketch over the input mesh to indicate the region for fitting and select the type of primitive to be fit. On top of this, we plan to crop the selected mesh data to generate a reduced mesh that encompasses only the area selected by the user. Finally, we utilize differentiable rendering techniques to automatically optimize the shape parameters of user-selected primitives to fit our reduced mesh data. With this tagging tool, we can enable more people without modeling expertise to tag objects. Data generated with this tool can support future research that studies object affordances with learning, as well as improve applications in robotics, product design, and assembly design like FabHacks.

Achieving consistent targeting of multiple simultaneous probes during electrophysiology experiments is a challenging and time-consuming process. Even with a planned insertion trajectory, experimenters still have to go through a lengthy process of positioning and inserting each probe. Electrophysiology experiments are increasingly focused on brain-wide coverage, requiring three or more simultaneous probes motivating researchers to accelerate their processes to reduce the duration of the experiment and the corresponding stress levels of their subjects. To improve the efficiency and reproducibility of multi-probe electrophysiology experiments, we developed two frameworks: a communication platform to allow software control of hardware micro-manipulators and an automation platform to perform multiple synchronous probe insertions. Each existing manipulator platform has proprietary software for programmatic control, which is rarely cross-platform and often exposes inconsistent interfaces. To standardize manipulator communication, we developed a Python server that acts as a generic cross-platform application programming interface (API). This platform ensures that client applications only need to interface with one API to be compatible with many different manipulator platforms connected across various computer operating systems. Building on top of this communication platform and an existing trajectory planning tool, Pinpoint, we next developed a system that automates the insertion process for multiple probes, saving time. The automation system provides three guarantees for researchers: first, that probes will reach their intended targets without manually introduced errors in targeting; second, that experiments can be repeated exactly to improve reproducibility; and third, that movement speeds are limited to low levels for reduced tissue damage. Because our software drives multiple probes simultaneously, complex multi-probe insertions are more manageable. Taken together, these open-source tools for communicating with hardware manipulators and automating multi-probe insertions enable the next generation of reproducible, high-efficiency, brain-wide electrophysiology data collection.

How much global warming will the Earth experience? This depends mostly on how quickly fossil fuels and other heat trapping gasses are phased out. We used reduced-complexity climate models to calculate whether a given emissions scenario meets temperature targets and other global effects. Our research starts with writing code that pulls and compiles the most recent data on various global environmental factors. This is used alongside existing data that break down emissions by industrial sectors, such as agriculture, electricity and transportation as well as by fuel, such as coal, oil, gas and land use. Using the updated historical data, we created various scenarios that ramp down emissions to zero over a specified number of years into the future. These scenarios were run through the Finite-amplitude Impulse-Response (FaIR) model to create plots demonstrating the resulting effect on global temperature. Additionally, we are considering the current decarbonization trends in our analysis. We noted current rates of decarbonization and continued these trends into the future to determine how much warming the earth will experience as a result. This data can be compared to the critical two degrees of global average temperature increase. By running these models, we can use current trends to estimate if we will exceed two degrees of global warming. Additionally, by modifying the rate of emission reduction, we can see what economic changes need to be made to stay under two degrees of global warming.

Sub-gram flying robots have transformative potential in applications from search and rescue, to precision agriculture, and environmental monitoring. A key gap in achieving autonomous flight for these applications is the low lift-to-weight ratio of flapping wing and quadrotor designs near 1 gram. To close this gap, we propose a helicopter-style design that minimizes size and weight by leveraging the high lift, reliability, and low-voltage of sub-gram motors. We take an important step to enable this goal by designing a light-weight, microfabricated flybar mechanism and tail wing rotor to passively stabilize the Coin-Copter: a helicopter-style robot. A 48 mg flybar is folded from a flat carbon fiber laminate into a 3D mechanism that couples the tilting of the flybar to a change in the angle of attack of the rotors. The Coin-Copter’s flybar uses a novel flexure joint design instead of the ball-in-socket joints common in larger flybars. This flybar achieved a peak damping ratio of 0.528, an 18.9x improvement from our initial design. Compared to a flybarless rotor with a near 0 damping ratio, our flybar-rotor mechanism can maintain a stable roll and pitch with relative deviations <1°. This research focuses on testing the yaw stability of a near-gram flying robot by incorporating and improving on flybar designs, roll-pitch-yaw test setups, and writing robot control software using pulse width modulation to help precisely control the heading of the Coin-Copter.

Fractal dimension, a measure of geometric complexity, finds application in image analysis, biology and medicine, neuroscience, geology and various other fields, yet existing methods often lack adaptability to finite data sets. Using ideas rooted in geometric measure theory, such as Hausdorff measure and Frostman’s Lemma, this research introduces a novel approach to compute fractal dimensions for finite sets, addressing limitations of traditional methods. Using Python, we developed and tested an algorithm to validate known sets such as the unit interval, square, cube, and fractal objects including the Cantor set and Sierpinski triangle. Comparative analysis was also conducted on established methods, including box-counting and correlation integral algorithms, to demonstrate the algorithm's accuracy in determining fractal dimensions. Pivoting towards data sets, we expect to use the computed fractal dimension of real data as a tool for assessing data and optimizing data compression. Our methods offer an improvement as most existing techniques use statistical methods that are limited to integer dimensions. In addition, recent studies have shown that fractal dimension values can be useful as features in machine learning. We also improve upon the calculation of the local dimension of regions in a data set, allowing for additional insights into complex data sets. This includes identifying regions of high complexity, and we expect to show that this allows for the more effective use of algorithms such as principal component analysis. All of these are increasingly important in our society due to the abundance of high-dimensional datasets in both the physical and social sciences. Overall, the benefits of studying novel ways of calculating the dimension of large data sets include efficient representation of data, improved interpretability, and decreased computational burden, as well as detecting certain features in data such as regions of high complexity.

Nilpotency degrees of finite-dimensional quadratic algebras carry essential information for their combinatorial and homological applications. It is known that the maximal nilpotency degree a finite-dimensional quadratic algebra with n generators can contain is at least n+1 for all n > 2. However, the optimality of this bound is still unknown. I propose a geometric visualization of the algebraic varieties of all quadratic algebras with n generators in degree d to find the true optimal bound. I then utilize this visualization to construct a linear program that deterministically determines whether a finite quadratic algebra with n generators exists that has a nilpotency degree of at least d. Thus far, I have verified that this algorithm will give the desired optimal bounds and have completed an implementation using Sage. I expect to find the true optimal bound on the maximal nilpotency degree of a finite-dimensional quadratic algebra with three generators shortly. However, the algorithm will require revisions for higher values of n due to scalability issues caused by its computational complexity. Knowing this optimal bound would solve several open problems in ring theory, including bounding the computational complexity of computing the global dimension of Koszul algebras. Finding a bound that extends to all algebras, including non-quadratic algebras, would also bound the computational complexity of determining if a finitely presented graded algebra is finite-dimensional

When creating video games, developers incorporate auditory components like music and sound effects which influence users’ gameplay experience. A game’s music is often designed with respect to the game’s context or plot, containing melodic and harmonic ideas that are continually developed. Existing research in ludomusicology and human-computer interaction have explored the role of music in these games, but few have considered what musical factors are the most easily perceived or most effective for conveying information. My work investigates specific elements of a game’s music, how they are perceived by a user, and how they impact the user’s decision-making. Participants complete a digital maze in which the music progressively adapts in response to their selected path but the adaptation method is not explicitly revealed to the user. Actions that bring a user closer or further to finishing the maze have opposing adaptations, though it is left to the user to observe and interpret these adaptations correctly. The adaptation methods include tempo, dynamic, pitch, and layering or texture. Through analyzing quantitative data tracked during gameplay as well as interviewing with participants about their experience, I seek out which of the aforementioned auditory changes are most easily perceived by and influential to players. I also discuss emotional responses associated with changes in certain auditory factors. Findings from this work may inform the development of software with effective and meaningful auditory elements for users.

Flying insect robots (FIRs), owing to their minuscule weight and size, offer unparalleled advantages in terms of material cost and scalability. However, their size introduces control hurdles, notably high-speed dynamics, restricted power, and payload capacities. While there have been notable advancements in developing lightweight sensors, often drawing inspiration from biological systems, the challenge remains in executing controlled flight without external feedback. We introduce Tiny Sense, a novel avionics system tailored for FIRs, encompassing an integrated sensor package — an inertial measurement unit, a pressure sensor, and an optical flow sensor. Coupled with a Kalman Filter (KF), this system weighs a mere 78.4 mg, drawing 15 mW of power. This is lighter and more power-efficient than previous sensor suites of the same capabilities. Our system uses a global-shutter camera as an optical flow sensor to collect pixel intensities for accurate optical flow calculations at 100 Hz. We collected raw data from the Tiny Sense by attaching it to a Crazyflie quadcopter and tested the KF by comparing its results to the measurements from the Crazyflie. We will continue to integrate the Tiny Sense with sub-gram FIRs and are currently working on mounting it to a 74-mg RoboFly. Our sensor suite allows even smaller FIRs to be able to achieve autonomous control.

Viscous Thread Printing (VTP) is a novel manufacturing technique that allows foam production using traditional Fused Deposition Modeling (FDM) printers. This printing technique takes advantage of an everyday phenomenon called Viscous Thread Instability (VTI), which can be observed when honey is drizzled onto pancakes. Similarly, molten filament buckles onto itself and creates a coiling pattern when extruded from enough height. These coils create cellular structures that have shown potential improved durability and expanded applications such as in medical scaffoldings. However, as a relatively new technique, VTP has been limited to producing single-stiffness (uniform density) foams in previous works, and it remained unproven whether we can produce VTP foams containing multiple densities. Drawing inspiration from biological structures with variable porosity such as bones and balsa woods, we hypothesized that we could create multi-density VTP foams by manipulating predominant VTP parameters that affected the size of the coils. This way, we can vary the pore sizes, and thus the density and stiffness, of a single cellular structure while preserving high structural integrity. Such structures would have many applications such as in robotics and prosthetics, such as customizable orthotics and limbs for soft robots. Beyond enabling this technique, we further investigate a novel methodology to simulate the printing process of variable density VTP foams and measure the foam's material properties. This allows for an easier and more sustainable exploration of the design of VTP foams without wasting any filament, which would make VTP foams more accessible in industry and research settings.

Geoparsing, the task of assigning coordinates to locations extracted from text, enables us to better understand how places change over time through historical documents. The task has seen many advancements with the growth of neural machine learning methods, but these models require large amounts of training data. In the history domain, many geoparsing corpora are from newspaper collections because of the availability of labeled data. However, not all historical research benefits from rich data availability, so we seek to understand and improve existing geoparsing methods for small corpora. In this project, I focus on the Svoboda Diaries, a collection of personal diaries written by British steamship purser Joseph Svoboda during the late 19th century in Ottoman Iraq. How do geoparsing methods perform in the Svoboda Diaries, a small historical text corpus written primarily in English, but also including text in Arabic, Italian, and French? We develop a map-based generate-and-rank approach with clustering of context words surrounding each location in the text. The location data is retrieved from two knowledge bases: GeoNames, a gazetteer with over 25 million geographical names, and Wikidata, a free and collaboratively edited database. We first extract the locations from the text using a natural language processing method called named entity recognition. Since modern location names may differ from historical spellings, we then generate alternate names for each identified location through translation and hand-crafted Romanization rules, such as “Basra” for “Basreh”, and query knowledge bases to retrieve coordinate data associated with these names. We find that our method to augment transliterated Arabic names with alternate names helps with gazetteer lookups, suggesting potential for augmenting data in other small corpora tasks. Our research with the Svoboda Diaries expands on existing geoparsing methods and can apply to other digital humanities projects facing the small corpora challenge.

Graph workloads are challenging for query optimizers in databases because of query features like larger sizes, frequent joins, and fewer filters. Traditional methods see large errors on queries with more joins, while machine learning methods tend to be complex and slower. We propose a framework to improve estimators by using graph colorings to make compact summaries of a data graph, storing important information about node relations. By modelling cardinality estimation as a subgraph matching problem, we can make use of this summary information and traverse the lifted graph to estimate the number of query graph matches. Additionally, we explore optimizations such as node summation and sampling to enable estimation even for larger queries. After evaluating various designs using this framework, we find improvements up to 100x to cardinality estimation accuracies compared to other recent methods while still maintaining efficient runtimes and memory usage. We discovered that quasi-stable colors, where nodes of one color have similar connections to other colors, result in these improved results when used to build the summary. These findings help improve graph database performance and offer a new application for graph theory.

The majority of people with Alzheimer’s dementia confirmed at autopsy are found to have one or more additional brain pathologies. To address this, we have developed a harmonized brain pathology score (BPS) across autopsy cohots that incorporates multiple forms of postmortem neuropathology. We sought to explore the genetic architecture of BPS using a systems-biology approach to further understanding of mixed pathology. We ran genome-wide association studies (GWAS) of BPS using HRC imputed data from European ancestry participants in each cohort separately, adjusting for age at death, sex, and population substructure. We performed meta-analysis using METAL. We performed gene-wide analysis using the GWAS results which we then integrated into the human protein-protein interaction (PPI) data using a dense module searching (DMS) method to identify network hub genes for BPS. We interrogated the Seattle Alzheimer's Disease Brain Cell Atlas (SEA-AD) dataset on the middle temporal gyrus to determine which cell types both hub genes were expressed in and how they differed across donors with higher degrees of AD pathology. The sample size consisted of 1,848 brain donors where 63% were females and mean age at death was 89.3. The quantile-quantile plot and genomic inflation (λ=1.005) for GWAS meta-analyses showed no bias. Apart from significant SNPs around the APOE region, we identified two candidate loci a) (Chr 9: rs1332179; MAF=0.1; P_meta=8.7×10-8) and b) (Chr 17: rs11078196; MAF=0.34; P_meta=1.9×10-7). The PPI network analysis identified VCP and IQCB1 as hub genes. While both hub genes were expressed broadly across cell types, IQCB1 was specifically higher with higher degrees of AD pathology in Microglia and VCP was lower with higher degrees of AD pathology in several neuronal populations in the SEA-AD dataset. Further functional enrichment analyses of these candidate loci are needed to determine whether these novel loci may identify targets for interventions to ameliorate AD.

Segmentation is an imaging technique commonly used to isolate an object of interest, such as an organ, from the background, or other objects in the image. When analyzing the shape of an anatomical structure, segmentation of that structure is often the first step in analysis. Precise anatomical segmentations are often created manually by subject experts, which is time-consuming, does not scale well, and can be prone to error since it is subjective. In this project, we aim to develop a machine-learning model to expedite whole-body surface segmentation from fetal mouse scans as part of an automated pipeline to detect asymmetry and abnormality in the facial region. The International Mouse Phenotyping Consortium (IMPC) has generated a large repository of three-dimensional (3D) imaging data from mouse embryos, providing a rich resource for investigating phenotype/genotype interactions. To generate segmentations required for training and validation of our deep learning model, the full body surface was manually segmented in 91 baseline scans from the IMPC’s Knockout Mouse Phenotyping Program (KOMP2) dataset. I trained a UNet with transformers (UNETR), on these segmentations that is able to estimate surface segmentations from new micro-CT mice images with an accuracy of 0.9. I am currently developing a fetal mouse full-body segmentation application powered by our deep learning model, SurfaceExtract, that will be made publicly available as an extension to the open-source image analysis platform, 3D Slicer. SurfaceExtract will be used by our lab to quickly and accurately generate segmentations of fetal mice as part of our lab’s automated facial asymmetry phenotyping pipeline.

As deep learning vision models become more prevalent, understanding the adversarial risk associated with them is important for maintaining safety and security. A common adversarial approach, evasion attacks, involve adding perturbations to the input data until it is correctly classified by humans but misclassified by machine learning models. Previous methods for physical-world evasion attacks include placing stickers, projecting artificial light sources, and casting shadows to mask the target object. The use of shadows, a naturally occurring phenomenon, is likely to remain undetected by people, and is therefore the focus of this project. Past shadow-based evasion attacks restrict the shadow design to more inconspicuous shapes, like triangles and other simple polygons. By designing a sculpture that can detract attention from the shadows it casts, this project aims to determine whether more complex shapes can be more successful at masking the target object. To compare the effectiveness of the shapes under the black-box setting, we use the same task as previous shadow-based evasion attacks, traffic sign classification, with the LISA and GTSRB datasets. To test the attack method in a simulated environment, I use SketchUp to create various sculpture designs that cast the selected 2D shapes. A model of the sculpture is then tested in a real-world setting, evaluating both general and scheduled attacks in indoor and outdoor environments. Because previous shadow-based evasion attacks are more effective when using polygons with more sides, we expect that complex shapes will result in a higher attack success rate.

As we receive spatial and temporal information, our brain develops sequential patterns to store events, giving us the ability to learn and store relationships. For learning and memory, these rapidly-encoded sequences are reactivated as “replay” sequences after experiencing the original trajectory, as often observed in the hippocampus. This brings up the question: what biological mechanisms enable us to build, encode, and trigger these relationships and replays? The goal of this project is to model sequential replay in spiking neural networks to explore and understand various biological mechanisms that produce the acquisition of such sequences. We are using NEST Simulator, a spiking neural network simulator software, to model large-scale neural networks. Then, we explore how changing dynamics such as non-random structure of the network and interactions between excitatory and inhibitory cells can contribute to sequence generation, as well as the salience and speed of such sequences. We have observed the significance of interplay between particular parameters, such as the widths of spatial Gaussian distributions for neuron connection strengths, by analyzing generated spiking raster plots. Recent work has also suggested an important role of long-term potentiation of intrinsic excitability in sequential replays, which we are integrating with the aforementioned dynamics by building a unique synapse model within the simulation software. This is a novel method to introduce excitability in a network, which is important to determine how changing excitability through potentiation, rather than plasticity, facilitates network formation and propagation. This research is significant because it highlights the components of neural networks that could be crucial to quickly generating and maintaining sequences for learning and memory, therefore helping us understand the brain’s mechanisms for storing spatiotemporal relationships.

The human proteome consists of tens of thousands of proteins produced from sequences translated from the human genome. Further, each of these proteins can be modified post-translationally to create an even larger set of unique proteoforms. With such a massive catalog, a tool that could accurately and inexpensively fingerprint proteins with single-molecule resolution in real-time would have a transformative impact on biology, medicine, and healthcare. To develop such an approach, we are utilizing nanopore sensor technology. Nanopores function by electrically-examining proteins at the molecular scale. As a protein molecule passes through the nanopore—a minuscule orifice in a thin membrane—it modifies the ionic current. Each protein induces a unique alteration in the current, producing a distinctive signal pattern, or 'squiggle'. These squiggles effectively act as molecular fingerprints, potentially enabling us to identify and classify different proteins based on their specific current changes as they traverse the nanopore. In this project, we are tasked with building a machine learning model to classify proteins based on their squiggle templates when passed through a nanopore. We found that a classification model based on a standard Convolutional Neural Network (CNN) performed well on simulated data of eight synthetic protein designs, but failed to generalize properly to experimental test data. In contrast, a 1-Nearest Neighbor model significantly outperformed the neural network architecture on the synthetic protein test data. We plan to assess this model's performance on fingerprinting the human proteome through a simulated dataset, with the ultimate goal of sharing our findings with other labs that specialize in developing precision medicine, targeted drugs, and technologies for understanding protein structure, function, and interaction.