Authors: Dylan Skinner
Mentors: Mark Hughes
Insitution: Brigham Young University
Deep reinforcement learning (DRL) continues to demonstrate remarkable efficacy in pattern recognition and problem-solving, particularly in domains where human intuition falls short. In the realm of knot theory, an important challenge revolves around constructing minimal-genus slice surfaces for knots of varying complexity. In this presentation, I will outline a novel approach that leverages the power of DRL to tackle this difficult problem. Using braid representations of knots, we train reinforcement learning agents to construct minimal genus slice surfaces by finding sequences of braid transformations that are optimal with respect to a given objective function. This provides a template for attacking other computationally difficult problems in topology and pure mathematics using reinforcement learning.