[00456] Network representations of attractors for surrogates generation and change detection
Session Time & Room : 2E (Aug.22, 17:40-19:20) @G709
Type : Contributed Talk
Abstract : Attractors arising from delay embedded time-series can characterise system dynamics. However, extracting useful representations is challenging for systems with high-dimensional or complex structure. We propose a data-driven method to represent attractors as networks, where dynamics are encoded as node transition probabilities. The usefulness of this representation is demonstrated in two tasks: (1) surrogate data generation; and (2) change point detection. These methods are applied to chaotic time-series, and experimental ECG data for heart attack detection.