This page contains machine learning projects that I have worked on.
Modeling Catan using self-play (2024)
Abstract: I taught a neural network how to play the board game Catan using reinforcement learning via self-play. When training, I utilized both temporal-difference and Monte-Carlo tree search methods, along with a residual neural network structure. The resulting model achieved an intermediate level of play.
The brain age project (2020)
Paper: Asher, Justin; Tan Dang, Khoa; and Masters, Maxwell (2020) "A Differential Geometry-based Machine Learning Algorithm for the Brain Age Problem," The Journal of Purdue Undergraduate Research: Vol. 10, Article 40. Link
Abstract: Our study aimed to find a new method for predicting the age of a subject from their brain biometrics. In order to measure the stretching of the brain tissue, we utilized an isoperimetric-type ratio $ T^2/A $, where $ T $ is the membrane thickness and A is the surface area, for each region of the brain. We found that a multinomial logistic model trained on the ratios preserved, and improved, prediction accuracy. This showed that our dimensionality reduction technique was effective for modeling brain age.