Inverse Reinforcement Learning with Uncertainty Quantification for Resource Selection in Collective Animal Systems

• Activity Number: 106 - AI and Deep Models for Spatial and Spatio-Temporal Data
• Type: Invited
• Date/Time: Monday, August 3, 2020 : 1:00 PM to 2:50 PM
• Sponsor: Section on Statistics and the Environment
• Author(s): Christopher Wikle* and Toryn Schafer
• Companies: University of Missouri and University of Missouri
• Keywords: deep learning; reinforcement learning; collective animal movement; individual-based models; spatio-temporal; ecology

Abstract

Collective animal movement is a complex spatio-temporal process due to complex decision making that involves interactions of animals with both environmental and social cues. The strength of the cues in driving behavior varies among life history strategies and can be difficult to learn. Individual-based methods allow one to define simple rules for actions that can generate complex group behaviors. However, traditionally it has been difficult to build flexible statistical models that can learn preferred rewards for resource selection from this perspective within a formal uncertainty quantification framework. Here, we use reinforcement learning to learn optimal rewards for decision-making that describe collective behavior with respect to resource selection. This is implemented in an inferential framework that allows uncertainty quantification.