Internship Location: West Long Branch, NJ
Organization Type: Private Sector
Policy Areas: Health policy
In my internship, I learned a lot about the intersection of machine learning and health policy. Reinforcement learning (RL) is a branch of machine learning that facilitates autonomous agents’ interactions with their environments. The agent’s main task involved navigating the hospital environment and picking up medicine, and then dropping the medicine off at a terminal location. The agent received a numeric reward for every action which, after several iterations, enabled it to learn the optimal path through the environment. The goal for this iteration of the project was to expand on the Q-Learning algorithm by instead using a more complex approach. This was accomplished by creating a more expansive environment which represented a more realistic hospital and transitioning the Q-learning algorithm to a deep Q-learning (DQN) algorithm. Instead of using Q-tables, DQN algorithms use neural networks. Q-tables are discrete, and only work effectively for simple problems with a limited number of actions an agent can perform. Neural networks, on the other hand, are continuous, and can work more efficiently with large quantities of possible actions that an agent can perform. Realistically, an agent navigating a hospital would be capable of performing many actions in a non-discrete environment. Therefore, switching to a DQN in a continuous environment provides a more realistic way to simulate a real-world implementation. The long-term goal of this project is to optimize the pickup and delivery of essential supplies and medications in hospitals, thus enhancing the efficiency of hospital operations and contributing to improved patient outcomes.