Bayesian inverse reinforcement learning for collective animal movement

Agent-based methods allow for defining simple rules that generate complex group behaviors. The governing of such models are typically set a priori, and parameters tuned from observed behavior trajectories. Instead making simplifying assumptions across all anticipated scenarios, inverse reinforcement learning provides inference on the short-term (local) long-term policies by using properties Markov decision process. We use computationally efficient linearly-solvable process to learn local collective movement simulation selfpropelled-particle (SPP) model data application captive guppy population. estimation behavioral costs is done in Bayesian framework with basis function smoothing. recover true SPP find guppies value more than targeted toward shelter.