Reinforcement learning for impulse stochastic control problems – ReLISCoP

This project is focused on building the mathematics to understand the design of optimal policy making and decision taking in complex socio-economic systems. By nature, these systems can only be controlled discretely and their evolution dynamics, which is complex, cannot be assessed perfectly. They also often exhibit delay in response to the action taken by the controller and possible random feedback effects. The ReLISCoP project aims at develop new impulse stochastic control models that are more realistic in term of applications, to propose a consistent mathematical study of them, to propose some reinforcement learning procedure in order to tackle uncertainty on models and, at the final step, to provide practical numerical tools that could be used in concrete situations.