Data-Driven Reconstruction of Strange Attractors in High-Dimensional Biological Signal Networks

Abstract

High-dimensional biological signal networks such as EEG, MEG, EMG, and multichannel physiological systems often evolve on latent nonlinear manifolds rather than in the full observation space. Recovering this hidden structure is important because biological states may differ not only in amplitude or frequency content but also in trajectory geometry. Recent studies on strange-attractor reconstruction and delay embedding show that complex temporal systems can be represented in reduced state spaces while preserving meaningful nonlinear organization. However, most biological applications still focus on isolated channels or extracted features rather than faithful reconstruction of attractor geometry in high-dimensional networks. This article develops a data-driven pipeline combining multichannel preprocessing, latent-coordinate extraction, delay embedding, manifold reduction, and nonlinear trajectory metrics. The results show that reconstructed attractors are bounded yet non-periodic, and that baseline, perturbed, and pathological states can be separated through differences in local divergence, correlation dimension, and attractor spread. Overall, the study shows that biological signal networks are interpreted more effectively through reconstructed latent dynamics than through conventional low-order summaries.