Reservoir Computing

Traditional machine learning and optimisation methods rely on digital handcrafted neural networks that are constantly increasing in size and complexity. Furthermore, they are complex to train, requiring billions of parameters and gigawatts of electricity. Conversely, reservoir computing uses the latent nonlinear components of dynamical "reservoirs" to train networks that are computational cheaper and easier to use. 

We are examining the use of various different kinds of reservoir computers and their applications in static and dynamic tasks with the domain of machine learning. These include mechanical, electromagnetic and photonic systems.