Emergence of a stochastic resonance in machine learning

TL;DR

This paper demonstrates that adding optimized noise to training data in reservoir computing can induce stochastic resonance, significantly improving the prediction accuracy and stability of chaotic systems.

Contribution

It introduces the concept of stochastic resonance in machine learning and shows how including noise amplitude as a hyperparameter enhances prediction performance.

Findings

Noise injection improves short-term and long-term prediction accuracy.

Including noise amplitude as a hyperparameter is crucial for inducing stochastic resonance.

The phenomenon is demonstrated on high-dimensional chaotic systems.

Abstract

Can noise be beneficial to machine-learning prediction of chaotic systems? Utilizing reservoir computers as a paradigm, we find that injecting noise to the training data can induce a stochastic resonance with significant benefits to both short-term prediction of the state variables and long-term prediction of the attractor of the system. A key to inducing the stochastic resonance is to include the amplitude of the noise in the set of hyperparameters for optimization. By so doing, the prediction accuracy, stability and horizon can be dramatically improved. The stochastic resonance phenomenon is demonstrated using two prototypical high-dimensional chaotic systems.