Reservoir computing: From self-organized dynamics to delay coupled hardware

This is an overview talk on recent developments on Reservoir computing. Reservoir computing originally introduced in the context of echo state or liquid state machines (LSM) has been proposed as a promising computational model for information processing in complex dynamic networks. Reservoir computing is a universal framework for computation, such as prediction, classification and memorization of information contained in time varying input streams. Here, I am going to present an overview of our research that extended the original LSM concept by incorporating self-organization based on neuronal plasticity. Further I will demonstrate that communication delays in recurrent system can help implementing super-fast and complex reservoir that can be understand by the follow same principles and that open up a new path towards super large neuromorphic hardware implementation. Thirdly I will include several demonstrators highlight the simplicity and powerfulness of this approach in real time control problems and for technological applications.