136879 research outputs found
Sort by
Fostering a Healthier Work Environment: An Artificial Intelligence-Based Approach for Identifying Individual and Team Wellbeing in Real-World Settings
No full textProfessionals with high job responsibilities, such as healthcare workers or astronauts, often experience significant stress due to their demanding roles. This stress can lead to burnout, causing anxiety and mental health issues. Team dynamics are crucial in these professions, and subtle behaviors can impact trust and team cohesion. Intelligent ambulatory tracking, using IoT devices like smartphones and wearables, offers personalized mood and emotion monitoring, especially relevant for mental health. This data supports tailored interventions to enhance work performance, reduce mental health problems, and improve both individual and team outcomes. Ambulatory data also provides a more objective alternative to self-reports, minimizing bias, and can be used for just-in-time adaptive interventions based on current states and contextual factors.
However, ambulatory monitoring faces several challenges. Ambulatory data are collected in uncontrolled real-world settings and they are susceptible to confounding factors like concurrent activities (e.g., exercise), interpersonal relationships (e.g., conflict with co-workers), and environmental conditions (e.g., indoor ventilation, and outdoor weather conditions). Ambulatory signals can also be affected by non-standardized factors like sensor placement and fluctuating physiological baselines, leading to highly variable data. Individual differences, such as demographics and psychological factors, impact human behavioral data across various signal types. Furthermore, the subtle, context-dependent nature of human behavior poses computational challenges in automatic detection, an area that has seen limited exploration. These challenges hinder the development of machine learning methods for reliably quantifying psychological and emotional outcomes from ambulatory data.
This dissertation addresses three main challenges through three studies: the high inter-individual variability observed in ambulatory data, the subtlety inherent in human behaviors, and the influence of context on human behavior. In the first study, we introduce personalized machine learning models for well-being detection. These models utilize a metric learning approach implemented through a Siamese neural network (SNN), focusing on relative patterns in personalized pairwise data rather than absolute patterns. Personalization is further achieved by incorporating individual traits as additional inputs and clustering criteria for group-based training of metric learning models. The approach is evaluated on healthcare professionals, detecting affect, stress, and anxiety from ambulatory multimodal data. The personalized models outperform baselines significantly (e.g., achieving a 3.77 times better correlation coefficient than the baseline with 90% of the target participants��� data used for fine-tuning).
In the second study, I explore ways to incorporate contextual information into machine learning models for detecting subtle interactions between individuals in real-life settings. Context is modeled at the conversational level by leveraging semantic connections between participants��� dialogue turns and integrating information about the task at hand and the underlying sentiment of the conversation. This context modeling is illustrated in the context of identifying "micro-behaviors" among team members, which are discreet indicators of thoughts and feelings toward one another. Analyzing longitudinal data from space simulations conducted at NASA���s Human Exploration Research Analog (HERA), the results indicate that including features from both participants significantly enhances micro-behavior detection performance compared to models utilizing features from each participant separately. The introduction of context in these models further improves micro-behavior detection, achieving an f1-score of 57.73%.
The third study delves into the influence of micro-behaviors on team performance using linguistic content analysis. The relationship between linguistic cues during these behaviors and team performance was evaluated on data same as the second study. This study also employed a deep learning model that merges a graph neural network (GNN) with a recurrent neural network (RNN) to assess team performance degradation based on language data. The GNN was used to model the interaction patterns among team members, while the RNN modeled the flow of conversation while handling tasks. The proposed model demonstrated superior performance compared to baseline models in an ablation study reaching a 56.16% F1 score, underscoring the significance of subtle team interactions in predicting team performance
Effectiveness of Geocell Reinforced Reclaimed Asphalt Pavement Base Layer for Flexible Pavement System over Expansive Subgrades
Get PDFThe transportation agencies must allocate a significant annual budget to rehabilitate low to high volume roads constructed over expansive subgrades. This type of subgrade soil undergoes substantial changes in volume due to the seasonal fluctuation of moisture levels, which will lead to pavement distress, including rutting, heaving, or/and longitudinal cracking. On the other hand, utilizing the large volume of reclaimed asphalt pavement (RAP) aggregates as a part of the pavement base layer has been a big challenge for researchers due to its poor mechanical properties. The traditional subgrade treatment procedures with full-depth reclamation are cost-intensive and time-consuming. Therefore, the transportation industry looks for an economical and sustainable alternative to address both these issues. This research study aims to assess the potential benefits of a three-dimensional confinement system, commercially known as ���geocell,��� to improve the performance of RAP materials and provide consistent support to the flexible pavement structure constructed over expansive subgrades.
This dissertation focused on contributing to the field of pavement geotechnics in two ways: first, to evaluate the performance of the geocell reinforced RAP-base (GRRB) layers to improve the performance of the flexible pavement constructed over expansive subgrade soil; and second, to develop a design methodology for such pavements based on field observations, cost and sustainability assessments.
Several test sections were constructed over an existing farm-to-market road, FM 1807, which suffered from distresses induced by the underlying expansive subgrade. These test sections were designed and constructed with different geocell-RAP infill materials, instrumented with sensors including Shape Array Accelerometers (SAAs) and Earth Pressure Cells. The structural capacities of the pavement sections were further evaluated by performing nondestructive field tests, including Falling Weight Deflectometer (FWD) and Automated Plate Load Test (APLT).
In addition to the field testing, numerical modeling analyses were performed to understand the contributions from the geocell bases and determine the future load-carrying capacity based on compressive strain acting on the subgrade soil. The expected design life of the geocell-reinforced pavement was calibrated with the field monitored data, and these results are used to develop flexible pavement design on expansive soils by utilizing a GRRB layer. The economic and sustainability aspects of flexible pavements with GRRBs are further verified with Life-Cycle Cost Analysis (LCCA) and sustainability analysis. It is believed that this research study will provide future practical guidelines for the construction and design of flexible pavements with GRRBs over expansive subgrades
De La Luna y Aqu��: Poems of Mysticism and Wonder
Get PDFPoetry is a liminal space where the metaphorical and the literal meet in equilibrium, and while this works for its own benefit in conveying profound emotion, poetry often can be lacking in terms of genre overlap. This dichotomy between the metaphorical and literal makes poetry a difficult medium for linear storytelling; difficult yet not impossible. Similarly, the genre of magical realism works to condense the metaphorical and literal into a palatable narrative not unlike a fairytale. Magical realism seeks to break down the supernatural into the natural and tangible.The intended goal of this project is to effectively combine both genres in order to explore the ways bi-ethnic cultures and identities within Latin-American immigrant families impact future generations. Indeed, magical realism as a means of expression has been mostly attributed to the novel as well as the visual arts. I find this medium sorely lacking within poetry and sought to research the potential the two have when in tandem together. Through this study in genre I aspire to investigate Latin-American perspectives on cultural identity. As the child of a Mexican immigrant who grew up surrounded by other children of similar heritage, I desire to create a work that reflects both the benefits and challenges of such an upbringing. This creative thesis will employ both genres and establish a comprehensive poetry collection that is a fictional narrative of a character named Victoria, who learns how to navigate these themes of culture, identity, family, and womanhood
Hypergraph Visualization With Terminal Based Interaction in Julia
No full textGraphs are an abstract data structure in which data points are represented by a collection of vertices (sometimes called nodes) and edges, in which the relation between any two vertices is shown by the presence, lack, or weight of an edge connecting them. Hypergraphs are an even more general data structure, in which these relationships can be more complex. Edges in a hypergraph, rather than always connecting exactly two vertices, may connect any number of vertices, allowing it to represent much more complex relationships between vertices. In research related to the utilization and advancement of hypergraphs and hypergraph theory, one hurdle is the difficulty and tediousness required to create a sensical visualization of a hypergraph. Due to the lack of tools made for this purpose, the majority of these visualizations will need to be made by hand. The software package resulting from this research project aims to fill this niche, being a command line, visually interactive program that can create and modify hypergraph components. A common representation of hypergraphs uses convex sets or convex hulls, which are simple polygons that enclose a set of points. In the context of hypergraphs, each point would represent a vertex in the hypergraph in two-dimensional space, and the convex hull would be the representation of the hyperedge, where all of the vertices within the convex hull would be members of that hyperedge, meaning that they are related to each other via whatever that hyperedge represents. One of the issues with this method is that it can occasionally generate false positives, which occurs when a node visually appears within a convex hull of a hyperedge that the node is not a member of. These visual false positives may lead to misinterpretations of the hypergraph. Another goal of this software package is the ability to identify when these false positives occur and notify its users
"Eliminating All Rapists from the Streets of Texas": A Case Study on Constitutional Carry and Domestic Violence Education Policies During the 87th Texas Legislative Session
No full textThis thesis project delves into the complexities of contradicting arguments regarding women���s safety in the 87th Texas Legislative Session, focusing on Constitutional Carry and Domestic Violence Education. I further determine how constructions of crime and public-private discussions impact policies which affect women disproportionately to men. I analyze how this translates to arguments made by conservative legislators in favor of Permitless Carry in Texas, citing concern for women, when the research suggests otherwise. Through the analysis of legislative discussions archived by the Texas Legislature, I conclude that women���s struggles are used to maintain a patriarchal power structure, even if subconsciously, which does not reflect the reality of crime against women accurately. Additionally feminist perspectives and women���s empowerment are appropriated to push a conservative agenda in Texas, with no consideration of the associated harm to women. Checking these theories against the outcomes of Senate Bill 1109, a domestic violence education piece, it is clear that any genuine concern for women���s safety is overshadowed by a desire to maintain patriarchal authority and traditional values