“Training Artificial Intelligence Agents to Play Games” by Maggie Haeussler, Lina Mo, and Sidney Wright,
Advisor: Dr. Stephenson

This talk explores the application of various artificial intelligence techniques in developing strategy for combinatorial games. We study a
family of 2-player combinatorial games played on m-by-n grids. Players alternate placing their pieces on the grid, and, when the grid is full,
sequences of pieces are scored based on length. Our research includes training AI agents with a variety of methods including tabular learning
and genetic algorithms involving artificial neural networks. We train these agents using a variety of non-learning agents and evaluate their
performance against both non-learning agents and one another to assess the quality of decision-making. Additionally, we study this family
of games theoretically, and we determine optimal strategies in many small cases.

“Modeling Long-Term Seed Viability with Survival Analysis: A 30-Year Study of Pioneer Plants in Costa Rican
Cloud Forests” by Wyatt Snyder, Advisors: Brian Yurk, Yew-Meng Koh

Pioneer plants play a crucial role in rainforest ecosystems. However, they grow from seeds that often wait significant time periods for canopy
openings before they can germinate. This makes the study of long-term seed viability especially important. We applied methods of survival
analysis to a seed viability experiment to model seed viability across time and determine the effects of microbial pathogen exposure on seed
viability for different pioneer species. Several parametric, semiparametric and nonparametric survival models were investigated, including a
novel parametric model incorporating initial viability. Models were fit to data from a 30-year seed viability experiment examining 6 pioneer
species native to the Cloud Forest of Monteverde, Costa Rica. The success of parametric models varied significantly from species to species
with some fitting well with a simple exponential or Weibull and others not fitting with any standard parametric model. Fit was significantly
enhanced by the addition of an initial viability parameter across many species and pathogen exposure treatment groups. These findings advance
the capabilities of parametric models to fit a greater variety of current-status data, particularly for ecological modeling.