hide
Free keywords:
-
Abstract:
The dynamics of highly parallel spike recordings can be characterized in terms of neural avalanches. The avalanche distributions reflect the correlations in activity across neurons, and across time. In vitro, avalanche size distributions f(s) were found to approximate power laws [1]: f(s)∼s−τ. This indicates diverging correlation lengths, and is characteristic for self-organized criticality (SOC) [2], prompting the hypothesis that neural population dynamics may be SOC. Moreover, SOC may provide a basis for optimal information processing [3]. However, evidence for the SOC hypothesis has been obtained only for coarse measures of neural activity (LFP, EEG, BOLD), but surprisingly not for spiking activity in vivo.
Therefore we analyzed highly parallel spike recordings from rats (hippocampus), cats (visual cortex) and monkeys (prefrontal cortex). Across all species, the avalanche size distributions f(s) were similar (Fig. 1A-C), but differed fundamentally from f(s) in critical spiking models (Fig. 1D), even when accounting for subsampling (Fig. 1E) (see [4–6]). The differences between in vivo f(s) and model f(s) could be overcome by decreasing the model’s excitatory synaptic strength, while increasing its external input (drive) concomitantly (Fig. 1F). Thereby the model became sub-critical and lost its separation of time scales (STS), which is fundamental to SOC. These results also held for intracranial depth recordings in humans.
Our findings indicate that the mammalian brain self-organizes to a slightly subcritical regime without STS. In this regime, avalanches are not temporally separated bursts as for SOC, but form a mélange. This regime may strike a balance between optimal information processing, which has been linked to SOC [3], and the need to avoid the supercritical regime, which has been linked to epilepsy.
