Pruning reservoir computing technology can reduce noise and reconstruct the nonlinear dynamics of potential engineering and neuroscience applications.
Many systems in the world are nonlinear. The change in the input is not proportional to the change in the output. However, scientists and engineers want to predict the behavior of these complex nonlinear systems in order to better understand and manage them. One challenge is to reconstruct the unobserved state from the limited and noisy measurements of nonlinear systems. Filtering techniques are often used to reduce noise, but they must include unknown equations governing the underlying physical process.
A promising option is reservoir computing, an artificial intelligence and machine learning technology that efficiently simulates systems using neural networks with random connections known as reservoirs. Sedehi et al. We present a truncated reservoir computing approach that distinguishes noise from signals and allows for reconstructing nonlinear dynamics from noiseless incomplete data without using physics-based models.
This approach is a form of data compression by pruning redundant nodes and edges, and increases the efficiency of the reservoir by optimizing hyperparameters with new machine learning protocols. The authors applied an approach to a simplified model of two nonlinear systems, Lorenz attractors and biological neuronal firing, to discover the ability to distinguish signals and noise to compete with traditional filtering techniques in low signal-to-noise ratios and high frequency ranges.
“This study paves the way for modelless removal and state reconstruction in complex systems with reservoir computing,” says author Omid Sedehi. “The proposed framework can be adapted to hardware implementations and is highly relevant to new applications of engineering, neuroscience, biological condition, and physical computation.”
Future work will focus on scaling pruning techniques to address high-dimensional reservoirs and provide hardware-enabled implementations for practical deployments.
sauce: “Removing and reconstructing nonlinear dynamics using truncated reservoir computing” by Omid Sedehi, Manish Yadav, Merten Stender, and Sebastian Oberst; chaos (2025). You can access the article https://doi.org/10.1063/5.0273505 .
This paper is part of the nonlinear dynamics of reservoir computing: theory, realization, application collection, details here .
