Learning Schooling Behavior from Observation

Agent-based simulation is a valuable tool for biologists studying animal behavior, however constructing models for simulation is often a time-consuming manual task, and validation of these models requires a principled approach. We present a framework for using machine learning techniques to automatically construct behaviors from tracking data of live animals from video that can be run in a simulated environment. Using this framework, we provide results for automatically learning the schooling behavior of Notemigonus crysoleucas.