Graduate Students Seminar

Speaker 1: Pratyusha Sarkar

Title

LOO-CV for high-dimensional Bayesian prediction models

Abstract

This project explores Bayesian leave-one-out cross-validation (LOO-CV) to estimate predictive accuracy. While LOO-CV is effective, it's computationally intensive. We address this using importance sampling, though traditional methods can suffer from infinite variance. To overcome this, we use a novel mixture estimator that maintains computational efficiency and ensures finite variance. Since individual parameter estimates μi^(mix) are dependent, we estimate the variance of the expected log predictive density (ELPD) using a batch means approach.

Speaker 2: Lara Scott

Title

Modeling Clustered Cell Migration During Oogenesis of Drosophila melanogaster

Abstract

Clustered cell migration is a process where a group of cells moves through biological tissue in response to a stimulus, and it plays a crucial role in wound healing, tissue development, and cancer metastasis. We study clustered cell migration using Drosophila egg chambers, leveraging imaging data to construct the complex geometry for a reaction-diffusion partial differential equation (PDE) for chemoattractant concentration in both two and three dimensions. We compare the two- and three-dimensional solutions and discuss how these solutions can be used to generate a migratory force for an agent-based model of the boundary points for each cell. To facilitate these simulations, we leverage the High Performance Computing Facility (HPCF) at UMBC.