This paper reports an approach to detecting the "early warnings" of upcoming global state transitions in a network based on its local dynamics, demonstrating that seemingly stochastic global events can be predicted by local deterministic dynamics. Based on the method using a nonlinear state-space reconstruction, we show that, surprisingly, dynamics of individual neurons can robustly predict the upcoming synchronous burst in the neural population at high signal-to-noise ratios, which even outperform the predictions based on population activity. We explain this apparently counterintuitive property by the network structures realizing in the nonbursting period, which is supported by a manipulative experiment and analyses. These results reveal basic properties of the bursting network dynamics.
Proceedings of the National Academy of Science
- Pub Date:
- September 2017
- Quantitative Biology - Neurons and Cognition;
- Mathematics - Dynamical Systems;
- Nonlinear Sciences - Chaotic Dynamics;