Reinforcement learning (RL) is a subfield of machine learning concerned with sequential decision making under uncertainty. In the past decade, RL has seen breakthroughs in game domains (such as AlphaGO and AlphaStar). However, applying RL to real-world applications is still challenging due to the requirement of online interaction and its susceptibility to distribution shift. Imagine trusting a health diagnosis to a system that might not be accurate if the microscope lenses are different from those used during training, or some other factor that causes the distribution of outcomes to shift.

In this team, students will investigate core RL methods that would enable real-world applications. Students will apply reinforcement learning to solve sequential decision making and combinatorial optimization problems encountered in healthcare and physical science problems, such as patient treatment recommendations using Electronic Health Records, molecular designs, and physical device designs.

Topics include model-based RL, offline RL, meta RL, and multi-objective RL. Also, we will develop reinforcement learning methods for healthcare and physical science applications.

Meeting time and location:
For academic credit, our MDP course is classified as a hybrid course but will mainly meet remotely, following university public health informed guidelines. Our team typically meets on Friday 3:30 pm –5:30 pm ET virtually. A best time will be finalized once students are identified. Each subteam arranges a convenient time to meet and work together. A two-term commitment will begin January 2021.

Team organization:
Each subteam has a team leader that reports to and meets with the faculty PI. The teams are flexibly structured to enhance creativity and opportunity for student growth.

More information