Item

Abstract

Criticality signatures, in the form of power-law distributed neuronal avalanches, have been widely measured in vitro and provide the foundation for the so-called critical brain hypothesis, which proposes that healthy neural circuits operate near a phase transition state with maximal information processing capabilities. Here, we revisit a recently published analysis on the occurrence of those signatures in the activity of a recurrent neural network model that self-organizes due to biologically inspired plasticity rules. Interestingly, the criticality signatures are input dependent: they transiently break down due to onset of random external input, but do not appear under repeating input sequences during learning tasks. Additionally, we show that an important information processing ability, the fading memory time scale, is improved when criticality signatures appear, potentially facilitating complex computations. Taken together, the results suggest that a combination of plasticity mechanisms that improves the network’s spatio-temporal learning abilities and memory time scale also yields power-law distributed neuronal avalanches under particular input conditions, thus suggesting a link between such abilities and avalanche criticality.