Machine learning is a tool that has become increasingly sought-after in data analysis, artificial intelligence, and software engineering, along with many other fields. One of the most pure forms of machine learning is reinforcement learning, which takes the simple concept of “rewards” for positive outcomes and “punishments” for negative outcomes, and transforms this into an algorithm for learning information. While the concept may seem basic, the mathematical theory behind reinforcement learning is quite extensive, as there are many different ways we can quantify “reward” and “punishment” in an algorithm.

This talk focuses on one specific branch of reinforcement learning called “Q-learning,” which makes use of a “Quality-Value” or “Q-value” function in order to keep track of positive and negative outcomes. We will test this algorithm on a game of sorts called “grid-world” in which an “agent” attempts to navigate a grid of numbers, searching for the highest possible reward. In analyzing the algorithm, it becomes apparent that we can also apply it to more “real-world” situations, such as finding the shortest path in a graph with edge-weights. These applications solidify the importance of Q-learning as a tool not just in math but in many other fields as well.