Speaker
Description
A properly designed controller helps to stabilize the system against environmental perturbations and improve the quality of experiment measurements. The task of tuning a controller to optimize its performance is the subject of optimal control theory and with development of deep reinforcement learning — a powerful new learning approach that employs artificial neural networks — the interest to optimal-control problems have been revitalized. Here we show how deep reinforcement learning method can be used to exert control on dynamics of stochastic systems. Specifically, we use agent-based simulations augmented with neural networks that act as a controller to explore two typical control strategies, namely, designing new dynamics or driving state-to-state transitions. We demonstrate the workflow and the effectiveness of the proposed control methods by considering the following three examples of stochastic processes: particle coalescence on a lattice, totally asymmetric exclusion process, and lattice active matter.