Random Reservoir Network
A neural network variant with a fixed randomly connected structure in which only the output layer is trained. It belongs to reservoir computing and is well suited to processing temporal signals.
Timeline (last 90 days)
A random reservoir network (often referring to an echo state network) is a type of recurrent neural network: the internal connection weights are randomly initialized and then kept fixed, and only the linear readout of the output layer is trained.
Random reservoir networks encode the short-term history of input sequences and the coupling among components through a high-dimensional dynamic state space, and combined with linear algebra or statistical tools they can extract strong signals of component composition and causal relationships.
Random reservoir networks have extremely low training cost and relatively strong interpretability, but their expressive power for long-term dependencies and discrete structures is limited.
All Facts (3)
A random reservoir network (often referring to an echo state network) is a type of recurrent neural network: the internal connection weights are randomly initialized and then kept fixed, and only the linear readout of the output layer is trained.
50%UnverifiedRandom reservoir networks encode the short-term history of input sequences and the coupling among components through a high-dimensional dynamic state space, and combined with linear algebra or statistical tools they can extract strong signals of component composition and causal relationships.
50%UnverifiedRandom reservoir networks have extremely low training cost and relatively strong interpretability, but their expressive power for long-term dependencies and discrete structures is limited.
50%