Combinatorics

Physics

Reservoir computing

Work (physics)

Mathematics

Machine learning

Recurrent neural network

Spectral radius

RADIUS

Artificial intelligence

Eigenvalues and eigenvectors

Interval (graph theory)

Computer science

Quantum mechanics

Computer security

Artificial neural network

Thermodynamics

Physical Review Research

This work reports on the emergence of an interval in the spectral radius of the neural network in which the prediction error is minimized. The phenomenon can be beneficial to the design of optimal reservoir computing, representing a step forward in understanding these machine-learning systems.

Model-free prediction of spatiotemporal dynamical systems with recurrent neural networks: Role of network spectral radius