Exploring Artificial Intelligence and Machine Learning
Introduction to Reinforcement Learning and Multi-Agent Systems (CSC 395) is an innovative course that explores reinforcement learning, which, according to Sutton and Barto (1992), “is learning what to do — how to map situations to actions — so as to maximize a numerical reward signal. The learner is not told which actions to take, as in most forms of machine learning, but instead must discover which actions yield the most reward by trying them.”
This in-depth introductory course is divided into two parts — the origins, foundations, and traditional algorithms of reinforcement learning; and its varied applications. The curriculum applies reinforcement learning to multi-agent systems (MAS) to provide connections to game theory and robotics, while also training students to conduct research and follow a scientific path.
“It was rewarding to see my students developing and discussing different versions of Q-learning — a popular reinforcement learning algorithm — and using the theory to play with parameters,” says Assistant Professor of Computer Science Fernanda Eliott. “I was thrilled to see that the course attracted so many students and delighted with the outcomes.”
The challenging curriculum will provide you with hands-on experience applying reinforcement learning, developing algorithms, and exploring different testbeds. You will learn the theory first, then gradually acquire the scientific tools for a group project that allows you to apply what you’ve learned through a poster presentation and research paper using LaTeX. You’ll develop highly transferrable scientific skills by asking investigative questions and analyzing the answers. According to Professor Eliott, class projects can be fantastic additions to a coding portfolio, and the paper can serve as a writing sample for students who go on to apply to grad school.
In her Ph.D. studies, Professor Eliott used reinforcement learning and moral reasoning to design and develop software architecture to explore decision-making and moral behavior. Her research interests, which focus on combining reinforcement learning with cognitive inspiration, inspired her to create this course focusing on traditional reinforcement learning and its applications to multi-agent systems.
“I wanted to offer students a flavor of [the] real challenges and accomplishments of developing computational approaches that use reinforcement learning in multiagent systems,” Professor Eliott says.