Meharry Medical College

The BioDIGS project is a nationwide initiative involving students, researchers and educators across more than 40 research and teaching institutions. Participants lead sample collection, computational analysis and results interpretation to understand the relationships between the soil microbiome, environment and health.

Neural NLP models are often miscalibrated, assigning high confidence to incorrect predictions, which undermines selective prediction and high-stakes deployment. Post-hoc calibration methods adjust output probabilities but leave internal computation unchanged, while ensemble and Bayesian approaches improve uncertainty at substantial training or storage cost. We propose UAT-LITE, an inference-time framework that makes self-attention uncertainty-aware using approximate Bayesian inference via Monte Carlo dropout in pretrained transformer classifiers. Token-level epistemic uncertainty is estimated from stochastic forward passes and used to modulate self-attention during contextualization, without modifying pretrained weights or training objectives. We additionally introduce a layerwise variance decomposition to diagnose how predictive uncertainty accumulates across transformer depth. Across the SQuAD 2.0 answerability, MNLI, and SST-2, UAT-LITE reduces Expected Calibration Error by approximately 20

BackgroundFormal compilations of implementation strategies like the Expert Recommendations for Implementing Change (ERIC; 1) are relevant for planning and evaluating implementation initiatives. The relationship between these strategies and the particulars of a specific initiative is not always clear, particularly when implementation occurs outside of organized healthcare settings. The present study reports on the process used to develop an adapted glossary of ERIC strategies for implementing evidence-based health promotion programs in African American churches.MethodsA glossary adaptation team composed of academics and community representatives met twice a month for six months to adapt implementation strategy definitions to fit the context of the present project. Collaborative discussions were held until consensus was reached for each strategy. This project-adapted glossary was then subjected to coding to document the types of changes that occurred during the adaptation process.ResultsThe glossary adaptation team collectively dedicated 99.5 person-hours to the meetings to obtain consensus for the strategies. The final strategy glossary retained 64 strategies relevant to the project, with 84.4% of strategies involving some level of adaptation. Most of the adaptation involved specifying project-specific actors, specifying that the innovation involved evidence-based health promotion programs, and noting the role the project's technical assistance team serves in supporting congregations during implementation.ConclusionsAdapting implementation strategy definitions to a specific project is a time-intensive process that challenges a team to carefully and creatively consider how strategies may be enacted in a specific context (e.g., faith-based communities). The consensus-based process also served as a type of cultural exchange between the project's academic partners, community partners, and consultants. The development of a project-specific glossary leverages the ability of project team members to employ implementation strategies as part of project planning and evaluation.

Sentiment analysis models exhibit complementary strengths, yet existing approaches lack a unified framework for effective integration. We present SentiFuse, a flexible and model-agnostic framework that integrates heterogeneous sentiment models through a standardization layer and multiple fusion strategies. Our approach supports decision-level fusion, feature-level fusion, and adaptive fusion, enabling systematic combination of diverse models. We conduct experiments on three large-scale social-media datasets: Crowdflower, GoEmotions, and Sentiment140. These experiments show that SentiFuse consistently outperforms individual models and naive ensembles. Feature-level fusion achieves the strongest overall effectiveness, yielding up to 4% absolute improvement in F1 score over the best individual model and simple averaging, while adaptive fusion enhances robustness on challenging cases such as negation, mixed emotions, and complex sentiment expressions. These results demonstrate that systematically leveraging model complementarity yields more accurate and reliable sentiment analysis across diverse datasets and text types.