Using Machine Learning to Optimize Can't Stop
Academic Level at Time of Presentation
Senior
Major
Mathematics/Data Science
List all Project Mentors & Advisor(s)
Christopher Mecklin, PhD
Presentation Format
Oral Presentation
Abstract/Description
Can’t Stop is a jeopardy dice game invented by Sid Jackson and first published in 1980. It has players rolling four dice to make pairs that advance markers up a board, with the first player to complete three columns winning the game. One of the most important decisions each turn is deciding whether to stop rolling the dice to keep your progress, or risk failing the next roll and losing your progress. Different strategies have been developed on when is the optimal time to stop, using a variety of statistical and mathematical approaches. However, one approach that has not seen significant development is neural networks, a machine learning method. Through experimentation and computer simulations, the goal is to see if using the reinforcement learning algorithm Deep Q-Learning (a type of neural network) will produce a better strategy for when to stop rolling than existing strategies, such as Rule of 28 (Keller 1986) or Generalized Heuristic (Glenn and Aloi 2009)
Spring Scholars Week 2026
Honors College Senior Thesis Presentations
Using Machine Learning to Optimize Can't Stop
Can’t Stop is a jeopardy dice game invented by Sid Jackson and first published in 1980. It has players rolling four dice to make pairs that advance markers up a board, with the first player to complete three columns winning the game. One of the most important decisions each turn is deciding whether to stop rolling the dice to keep your progress, or risk failing the next roll and losing your progress. Different strategies have been developed on when is the optimal time to stop, using a variety of statistical and mathematical approaches. However, one approach that has not seen significant development is neural networks, a machine learning method. Through experimentation and computer simulations, the goal is to see if using the reinforcement learning algorithm Deep Q-Learning (a type of neural network) will produce a better strategy for when to stop rolling than existing strategies, such as Rule of 28 (Keller 1986) or Generalized Heuristic (Glenn and Aloi 2009)