Echo-state networks

tags

Recurrent neural networks, Unsupervised learning

resources

Scholarpedia

Principle

An echo state network is usually a standard RNN with fixed random weights. The output from this RNN is used as a high dimensional feature map to be fed into a machine learning system.

(Jaeger 2004; Jaeger et al. 2007; Jaeger 2012)

Bibliography

1. H. Jaeger. April 2, 2004. "Harnessing Nonlinearity: Predicting Chaotic Systems and Saving Energy in Wireless Communication". Science 304 (5667):78–80. DOI.
2. Herbert Jaeger, Wolfgang Maass, Jose Principe. April 2007. "Special Issue on Echo State Networks and Liquid State Machines". Neural Networks 20 (3):287–89. DOI.
3. Herbert Jaeger. February 1, 2012. "Long Short-term Memory in Echo State Networks: Details of a Simulation Study". Jacobs University Bremen.