Takens-inspired neuromorphic processor: A downsizing tool for random recurrent neural networks via feature extraction

This article proposes an approach that uncovers some relevant features that are being created inside random recurrent neural networks' spaces, and that are fundamental for their prediction capabilities. The characteristics of the networks are used to downsize the amount of neurons that make up typical hidden layers of artificial neural networks. The authors further propose a hybrid Takens-inspired neuromorphic concept, in which the network is extended by a set of virtual neurons.