Research summary
An adaptive “autonomous agent” is a software program that responds to events in its environment without needing prompts or instruction from the user or owner. As Canada Research Chair in Reinforcement Learning, Dr. Martha White aims to develop techniques to help adaptive autonomous agents learn and adapt.
For example, consider an agent whose goal is to improve the efficiency of a water treatment plant. One of the primary costs is in the operation of mixers. The agent constantly makes decisions about mixing speeds and receives feedback about energy use and clean water production rates. Ideally, the agent should be able to learn how to adjust mixing speeds with minimal input from the user. It should also be able to continually learn (and improve) during the operation. White and her research team are developing prediction and control systems like this hypothetical one that can learn and adapt while in use.