Reservoir computing with higher-order interactive coupled pendulums.

Area of Science:

• Complex Systems
• Computational Neuroscience
• Nonlinear Dynamics

Background:

• Reservoir computing uses dynamical systems for time series analysis.
• Current methods rely on random matrices, leading to extensive hyperparameter tuning.

Purpose of the Study:

• Propose a novel, nonlocally coupled pendulum model as a reservoir computing architecture.
• Reduce hyperparameters and eliminate reliance on random matrices.
• Simplify hyperparameter optimization for complex system modeling.

Main Methods:

• Developed a nonlocally coupled pendulum model with higher-order interactions.
• Employed Bayesian optimization for efficient hyperparameter exploration.
• Trained the model to reproduce chaotic attractors using Lorenz and Hindmarsh-Rose systems.
• Analyzed prediction performance using Pearson correlation and Hausdorff metrics.

Main Results:

• The pendulum model successfully reproduced chaotic attractors with fewer hyperparameters.
• Higher-order interactions were shown to improve prediction performance.
• The chimera state was identified as the optimal dynamical regime for prediction.
• Effectiveness validated on Lorenz and Hindmarsh-Rose systems.

Conclusions:

• The proposed pendulum-based reservoir offers a more efficient and effective alternative to traditional methods.
• This novel reservoir structure has potential applications in physical system dynamics modeling.
• The findings highlight the importance of higher-order interactions and specific dynamical regimes for accurate predictions.