Multidisciplinary Institute in Artificial intelligence

LEVERAGING OFFLINE DATA TO CALIBRATE ONLINE REINFORCEMENT LEARNING AGENTS

ABSTRACT

Reinforcement Learning provides a general formalism to address control problems, but has yet to be widely used in the real-world. We have been investigating using reinforcement learning to optimize drinking water treatment, directly on the physical system, and have naturally run into many technical challenges. In this talk, I will discuss one such important question for real-world deployment: selecting hyperparameters offline, for a continual learning agent. Before a learning agent can be deployed, its hyperparameters need to be specified. I will outline one potential solution, under the setting where an offline batch of data is available from a previous controller in the system, and discuss key open challenges.

 MARTHA WHITE

Martha White is an Associate Professor of Computing Science at the University of Alberta and a PI of Amii–the Alberta Machine Intelligence Institute–which is one of the top machine learning centres in the world. She holds a Canada CIFAR AI Chair and received IEEE’s “AIs 10 to Watch: The Future of AI” award in 2020. She has authored more than 50 papers in top journals and conferences. Martha is an associate editor for JMLR and TPAMI, and has served as co-program chair for ICLR and area chair for many conferences in AI and ML, including ICML, NeurIPS, AAAI and IJCAI.