By Davide Conte (University of Campania “Luigi Vanvitelli”, Caserta, Italy). November 7th, 2024.
Understanding criticality in neural networks is essential for deciphering brain function and detecting pathological deviations. However, neural recordings often capture only a small fraction of the system, meaning that the local, subsampled behaviour may not accurately reflect global dynamics. In this study, we investigate two stochastic models, the mean field branching process and the (2+1)-dimensional directed percolation, to assess how subsampling affects power-law distributions. We find that while avalanche size and duration distributions are significantly altered, the exponents governing the power spectrum and detrended fluctuation analysis remain invariant across certain frequency ranges. The Crackling Noise relation, that connects the exponents of the avalanche distribution to those of power spectrum and DFA, is indeed violated in subsampled systems, due to the correlation that exists between different avalanches, that are actually fragments of a single large avalanche in the whole unobserved system. The invariance of power spectrum and DFA exponents represent another method for predicting the exponents of the global system from subsampled data, in a simple and unbiased way, offering a more accurate framework for inferring brain-wide critical dynamics and identifying early deviations that may signal neurological disorders.
Contact: Davide Conte, davide.conte13@gmail.com
Additional authors: Antonio de Candia, Department of Physics “E. Pancini” University of Naples Federico II, decandia@unina.it