Reservoir computing with large valid prediction time for the Lorenz system
Fuente:
arXiv
Salvato in:
| Autori principali: | , |
|---|---|
| Natura: | Preprint |
| Pubblicazione: |
2025
|
| Soggetti: | |
| Accesso online: | |
| Tags: |
Aggiungi Tag
Nessun Tag, puoi essere il primo ad aggiungerne!!
|
| _version_ | 1866916889951731712 |
|---|---|
| author | Hurley, Lauren A Shaheen, Sean E |
| author_facet | Hurley, Lauren A Shaheen, Sean E |
| contents | We study the dependence of the Valid Prediction Time (VPT) of Reservoir Computers (RCs) on hyperparameters including the regularization coefficient, reservoir size, and spectral radius. Under carefully chosen conditions, the RC can achieve approximately 70% of a benchmark performance, based on the output of a single prediction step used as initial conditions for the Lorenz equations. We report high VPT values (>30 Lyapunov times), as we are predicting a noiseless system where overfitting can be beneficial. While these conditions may not hold for noisy systems, they could still be useful for real-world applications with limited noise. Furthermore, utilizing knowledge of the Lyapunov exponent, we find that the VPT can be predicted by the error in the first few prediction steps, offering a computationally efficient evaluation method. We emphasize the importance of the numerical solver used to generate the Lorenz dataset and define a Valid Ground Truth Time (VGTT), during which the outputs of several common solvers agree. A VPT exceeding the VGTT is not meaningful, as a different solver could produce a different result. Lastly, we identify two spectral radius regimes that achieve large VPT: a small radius near zero, resulting in simple but stable operation, and a larger radius operating at the "edge of chaos." |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2508_06730 |
| institution | arXiv |
| publishDate | 2025 |
| record_format | arxiv |
| spellingShingle | Reservoir computing with large valid prediction time for the Lorenz system Hurley, Lauren A Shaheen, Sean E Neural and Evolutionary Computing We study the dependence of the Valid Prediction Time (VPT) of Reservoir Computers (RCs) on hyperparameters including the regularization coefficient, reservoir size, and spectral radius. Under carefully chosen conditions, the RC can achieve approximately 70% of a benchmark performance, based on the output of a single prediction step used as initial conditions for the Lorenz equations. We report high VPT values (>30 Lyapunov times), as we are predicting a noiseless system where overfitting can be beneficial. While these conditions may not hold for noisy systems, they could still be useful for real-world applications with limited noise. Furthermore, utilizing knowledge of the Lyapunov exponent, we find that the VPT can be predicted by the error in the first few prediction steps, offering a computationally efficient evaluation method. We emphasize the importance of the numerical solver used to generate the Lorenz dataset and define a Valid Ground Truth Time (VGTT), during which the outputs of several common solvers agree. A VPT exceeding the VGTT is not meaningful, as a different solver could produce a different result. Lastly, we identify two spectral radius regimes that achieve large VPT: a small radius near zero, resulting in simple but stable operation, and a larger radius operating at the "edge of chaos." |
| title | Reservoir computing with large valid prediction time for the Lorenz system |
| topic | Neural and Evolutionary Computing |
| url | https://arxiv.org/abs/2508.06730 |