NeurIPS 2019 Expo Panel
Dec. 5, 2020
Explainable Reinforcement Learning
Oliver Schulte (Simon Fraser University, Sportlogiq), Norm Ferns (Sportlogiq)
Oliver Schulte (Simon Fraser University, Sportlogiq)
This panel will discuss Explainable AI (XAI) challenges unique to Reinforcement learning (RL), and possible solutions to them. Our panelists are prominent researchers with in-depth industrial experience.
XAI refers to the application of AI in such a manner that any automated process or output can be well-understood and trusted by a human expert. RL is a sequential decision-making framework for modelling agents who learn to act in an uncertain environment. Learning is based on environmental feedback in the form of a numerical control signal intended to direct desired behaviour, in contrast to other machine learning methods that explicitly label correct / incorrect behaviours. Robust industrial application has found sporadic success, primarily in decision support, e.g. expert recommendation in healthcare, media, etc. In order to facilitate wider industrial adoption of RL techniques, it is vital that RL models be easily explained in the language of clients and domain experts.