Reservoirs as Temporal Filters and Feature Mappings

摘要

Parametrized state space models in the form of recurrent networks are often used in machine learning to learn from data streams exhibiting temporal dependencies. To break the black box nature of such models it is important to understand the dynamical features of the input driving time series that are formed in the state space. I will talk about a framework for rigorous analysis of such state representations in vanishing memory state space models, such as echo state networks (ESN). In particular, we will view the state space as a temporal feature space and the readout mapping from the state space as a kernel machine operating in that feature space. This viewpoint leads several rather surprising results linking the structure of the reservoir coupling matrix with properties of the dynamic feature space.