Power laws, memory capacity, and self-tuned critical branching in an LIF model with binary synapses

摘要

Both fluctuations and distributions of spontaneous neural spiking activity have been observed to closely follow a variety of power laws. Multiple explanations have been offered for each observation, but few lead to mechanisms that encompass their widespread occurrence. A canonical, leaky integrate-and-fire model is presented in which synapses are updated based on the timing of pre- and post-synaptic spikes in order to maintain a state of critical branching. Results showed that 1) the self-tuning algorithm maintained critical branching under a range of parameters; 2) power laws were obtained in spiking activity fluctuations (1/f scaling), size distributions of network bursts (neural avalanches), and temporal correlations in interspike intervals (Allan factor); 3) power laws disappeared once the self-tuning algorithm was disabled; and 4) critical branching was adaptive in that it maximized the network’s memory capacity when assessed as a liquid state machine.