82837 research outputs found
Sort by
The role of taz in hindbrain morphogenesis
No full textits development.
In this thesis, I describe a novel role and regulatory mechanism for Taz, a co-transcriptional
regulator initially discovered as part of the Hippo signalling pathway, during hindbrain and
hindbrain ventricle morphogenesis using the zebrafish model. Previous work found that loss of
Taz results in a smaller hindbrain ventricle and failure of midline separation. In addition,
immunofluorescence analyses show that apicobasal polarity components and the cytoskeletal
structure (via F-actin) within the hindbrain ventricle are disorganized. Furthermore, Taz is
localized to the boundaries of compartments in the hindbrain called rhombomeres (corroborated
by other work). I have demonstrated via pharmacological manipulation that this localization is
dependent on Wnt activity, where in a Wnt-OFF state, less Taz is detected in the hindbrain.
Conversely, in a Wnt-ON state, Taz is detected both at the boundaries and within the
rhombomere. Thus, demonstrating an in vivo Wnt-Taz interaction in zebrafish
Additionally, I observed that in taz -/- mutants, known boundary markers such as rnfg are
disorganized. Therefore, I sought to probe other genes that are expressed at boundaries. Using in
situ hybridization, my work has identified a series of other genes expressed in the hindbrain
either at the boundaries or immediately juxtaposed are also disorganized, including rasgef1ba,
rac3b, metrnla, and delta, suggesting that the loss of Taz affects boundary cell expression.
Furthermore, I probed other patterned genes such as the proneural progenitor markers ascl1b,
neurog1 lhx2b, and lhx9 in taz-/- mutants and saw mild changes. Analyses of boundary formation
and maintenance through egr2b in situ hybridization and changes to boundary cell shape show
that boundary cells are initiated normally and are functional.
Overall, this thesis aims to investigate the role of Taz in the hindbrain, from its upstream
regulators to downstream effectors. From this work I propose a model in which Taz regulates
boundary cell identity and is necessary for ventricle development
Item Revision with Network Psychometrics and Natural Language Processing
Get PDFConstruct-irrelevant variance (CIV), defined as excessive variance that is unrelated to the intended construct, poses a significant threat to the validity of test interpretations and applications. CIV can arise from two notable sources: construct-irrelevant items, which include items with content unrelated to the construct being measured, and redundant items, which repeat information already captured by other items, leading to individual fatigue and inflated reliability estimates. Consequently, test developers typically revise these items for future implementations. However, these solutions predominantly focus on the statistical aspects of the survey, leaving the linguistic aspect to mere observation by item writers.
This dissertation combines the capability of network psychometrics – a method that examines structure of a psychological instrument alternative to the traditional psychometric methods such as item analysis or factor analysis, and natural language processing (NLP) to identify and revise CIV items in the HEXACO personality inventory dataset. Network psychometrics results were validated with results from the traditional psychometric methods. Additionally, items revised by NLP were validated by cognitive rating surveys with both non-expert and expert population in the personality testing area.
This dissertation demonstrates a novel combination of psychometrics techniques and NLP to examine and improve a psychological instrument through the identification of CIV items. Theoretically, the proposed recommendation may serve as guidance for test developers on how to incorporate insights from psychometrics analysis and NLP into the item revision process. From a methodological perspective, this dissertation demonstrates the leveraging of item content data through the application of NLP techniques
Supermask-based Federated Learning Under Poisoning Threats
No full textFederated Learning (FL) has become an important paradigm for decentralized model training, enabling several users to work together to enhance a global model while maintaining the privacy of their personal information. FL has advantages, but it is also susceptible to poisoning attacks, in which adversaries intentionally introduce malicious updates that harm the global model. Conventional strategies for preventing this kind of attack have centered on a variety of robust aggregation techniques, but they frequently fail to strike a balance between robustness against attacks and preserving model accuracy.
To overcome these difficulties, the Federated Rank Learning (FRL) method was developed, which uses rank-based aggregation to make the global model robust against potentially dangerous updates. Although FRL has the potential to reduce the consequences of poisoning, there are certain drawbacks, especially when dealing with complex or widespread attacks.
In this work, we propose a novel enhancement to the FRL framework by integrating the Edge-popup algorithm with median aggregation. The Edge-popup algorithm, initially developed for neural architecture search, improves model robustness by selectively activating the most relevant weights of the network. Our method seeks to improve the robustness of federated learning systems against poisoning attacks by combining this algorithm with median aggregation, which is less susceptible to outliers.
We perform comprehensive experiments to evaluate the performance of our algorithm under different scenarios. Our results show that our method outperforms the existing methods in both robustness and accuracy
The Influence of Artificial Intelligence Agents’ Conversational Content on Consumers
No full textAI agents are increasingly employed to administer online chat customer service. This research examines the impact of AI agents’ conversational design on consumers. Specifically, I focus on AI chat agents’ conversational content in service interactions and test how consumers respond to their use of three types of conversational content: informational (e.g., what is your order number?), normative (e.g., please wait), and relational (e.g., how are you?). I propose that AI agents’ use of specific conversational content may have positive or negative effects on service outcomes depending on how authentically these contents represent AI agents’ true abilities. In particular, I show that, compared to using only informational content, when AI agents use informational and normative content, consumers perceive the agent as a more authentic AI, which leads to more positive service outcomes. However, compared to using only informational and normative content, when AI agents use informational, normative, and relational content, consumers perceive the agent as a less authentic AI, which leads to less positive service outcomes. I also explore two moderators of the proposed effects: the presence of anthropomorphic cues (i.e., humanlike avatar and name) and the service interaction outcome (i.e., success, failure). This research contributes to the literature by providing a systematic examination of AI chat agents’ conversational design, a fundamental aspect of AI design in text-based online chat interactions. This research also offers practical implications for AI and conversational designers, marketers, and firms by providing clear guidelines to firms on how to design their AI chat agents to appear more authentic and boost satisfaction
Epidemiology of Opioid Use Disorders in Critically Ill Patients at the Royal Alexandra Hospital, Edmonton – An Observational Cohort Study
No full textBACKGROUND: The precipitants of the ongoing opioid crisis are multifactorial and have evolved to become a critical public health challenge, contributing to increased rates of emergency department visits and hospitalizations due to overdose, opioid use complications, and issues with opioid withdrawal. This study evaluates the sociodemographic and clinical characteristics, as well as patient-centered outcomes and health services use among patients admitted to the intensive care unit (ICU) with known or presumed opioid use disorder (OUD). This study further aims to understand current gaps in knowledge related to OUD among patients admitted to ICU and possibly inform new therapeutic approaches or strategies.
METHODS: I conducted a retrospective observational cohort study of adult patients (≥18 years) admitted to the mixed medical/surgical ICU at the Royal Alexandra Hospital (RAH) in Edmonton, Alberta, between December 13, 2021, and December 13, 2022. The primary exposure was the presence of pre-existing known or presumptive OUD, defined using the Diagnostic and Statistical Manual of Mental Disorders, Fifth Edition (DSM-V) criteria abstracted though chart audit and adjudicated by ICU attending clinicians. Data were extracted from eCritical/TRACER Alberta, electronic medical records, and entered using standardized case report forms into a customized REDCap database. Data were captured on the sociodemographic descriptors (e.g., age, sex, and housing instability), clinical characteristics (Clinical Frailty Scale [CFS] scores, Acute Physiology and Chronic Health Evaluation [APACHE] II and III scores, admission diagnostic categories). The primary patient centered outcomes included mortality in ICU and hospital. The primary health services use outcomes included lengths of stay in ICU and hospital. Data are presented using descriptive statistics and compared across exposures, covariates and outcomes using Wilcoxon rank-sum tests, Chi-square tests, and Mantel-Haenszel tests, respectively.
RESULTS: In total, 1,089 patients were included, of which 63 (5.8%) were determined to have OUD. In total, the mean (SD) age was 57.1 (16.8) years, 41.7% (n=454) were female, and the median (IQR) CFS score was 4 (3-5). The mean (SD) APACHE II and III scores were 20.7 (8.8) and 69.9 (31.3), respectively. In total, 81.9% (n=695) were admitted with medical (non-surgical) diagnoses. Of the 63 patients with OUD, 48.3% (n=29) reported housing instability. Patients with OUD were younger compared to patients without OUD (median [IQR] age 37 [30-46] years vs. 61 years [48-70], p<0.01). CFS scores were lower among patients with OUD compared to those without OUD (median [IQR] 3 (2-4) for OUD vs. 4 (3-5) for non-OUD, p=0.007). Mean APACHE II (19.1 [7.0] vs. 20.8 [8.9], p=0.235) and APACHE III scores were not significantly different between those with and those without OUD Medical (non-surgical) indications were the predominant indication for ICU admission (89.8% for OUD vs. 81.8% for non-OUD, p=0.003). Patients with OUD were less likely to have undergone acute surgical interventions. ICU mortality was numerically but not statistically higher for patients with OUD compared to patients without OUD (22.2% [n=14/63] vs. 15.6% [n=160/1026]; odds ratio (OR) 1.55; 95% CI, 0.84-2.85; p=0.163). Hospital mortality was significantly higher for patients with OUD compared to patients without OUD (44.4% [n=28/63]) vs 31.5% [n=323/1026]); OR 1.74; 95% CI, 1.05-2.90; p=0.04). The ICU length of stay (LOS) was not significantly different between groups (median [IQR] ICU LOS 3.4 (2.2-5.6) days for OUD vs. 3.4 (1.7-6.5) days for non-OUD, p=0.945). The hospital length of stay (LOS) was significantly lower in patients with OUD compared with non-OUD (median [IQR] was 4.8 (2.4-8.0) days for OUD vs. 10.0 (4.4-19.7) days for non-OUD, p<0.0001).
LIMITATIONS: This study has limitations. First, this study was observational in design, therefore causal inferences are not possible. Second, this study was performed in a single urban hospital ICU, therefore limiting generalizability to other jurisdictions. Third, the determination of OUD (the primary exposure) was challenging to operationalize at the bedside and may have resulted in misclassification bias; however, it is more likely that patients with OUD were missed leading to an underestimate of the true proportion of patients admitted to ICU with OUD. Fourth, selected variables in the dataset had a high rate of missingness (e.g., data for CFS scores was missing for 26.6%), therefore potentially introducing risk of information bias. These data were not replaced or imputed. Finally, the study focused on short-term hospital-based outcomes and did not capture data on longer term outcomes among survivors after hospitalization.
CONCLUSION: Patients admitted to ICU with OUD were found to be younger, less likely frail, with comparable acuity of illness compared with patients without OUD. A large proportion of patients with OUD also experienced structure inequities, particularly insecure housing. Patient with OUD experienced higher mortality in ICU (non-significant) and hospital, underscoring this cohort with OUD are vulnerable and may benefit from multi-disciplinary care processes and transitions of care following acute care hospitalization
Teachers’ Designedly Incomplete Utterances in Chinese-as-a-Second-Language Classrooms
Get PDFThis study examines the multimodal resources used in Designedly Incomplete Utterances (DIUs) by teachers and students’ responses to these DIUs in Chinese-as-a-Second-Language (CSL) classroom interaction. Drawing on Conversation Analysis, Interactional Linguistics, and Multimodal Analysis, it focuses on syntactic, prosodic, and bodily-visual resources utilized in or alongside DIUs. By analyzing 312 instances of DIUs, the study reveals how DIUs used at different sequential positions are multimodally constructed.
The findings indicate that DIUs appear in three sequential positions—initiation, post-initiation, and post-response—each marked by syntactic forms exhibiting varying degrees of completeness and projection. Teachers use DIUs to elicit either structured responses (e.g., vocabulary drills) or open-ended, student-generated responses, in alignment with the pedagogical goals of the activity. The study also identifies post-response DIUs as repair devices, showing how syntactic resources assist students in identifying and correcting errors based on error types. Mandarin DIUs display syntactic features distinct from those in English, highlighting language-specific resources for managing morpho-syntactic incompleteness, especially in right-omission repairs.
Prosodic features play an important role as well. The analyses in Chapter 5 show that the final syllables of DIUs vary in pitch, register, and duration, shaped by the tonal nature of Mandarin and the participation framework in L2 classroom activities.
This dissertation further investigates bodily-visual resources accompanying DIUs, such as gaze direction, bodily holds, hand gestures, and visual displays. Gaze, for instance, is used to manage student participation and reference course materials, while bodily holds and hand gestures function as pre-completion signals, reinforcing the instructional context and aiding students’ anticipation to responses. Notably, pointing gestures and visual displays (e.g., character writing on blackboards or images on slides) are employed to link visual signs to pedagogical objectives. These multimodal practices highlight the integration of pre-class lesson planning with classroom interaction, emphasizing how visual materials, layout, and technology choices shape classroom dynamics.
This study enhances our understanding of how teachers in CSL contexts use multimodal resources to manage classroom interaction and engage students in participation. It underscores the importance of incorporating multimodal analysis into language pedagogy and highlights the need for lesson planning that integrates both linguistic and visual semiotic resources to enrich CSL learning