Abstract
The brain is a highly recurrent, nonlinear network hypothesized to remain near the edge of chaos for optimal performance. Excitation and inhibition must be balanced precisely within every neuron to ensure a consistent level of dynamical stability and rich dynamics during transition to chaos. However, analysis of biologically realistic synaptic weight matrices suggests that sparsity and low-dimensional structure interact such that there is no known synaptic balancing rule that constrains the stability (i.e., eigenvalues) of the network while also preserving computationally useful, low-dimensional structure. Further, even if a network were well-balanced, external stimuli interact with the nonlinear activation functions to unbalance the network in real time. Therefore, the brain must utilize dynamic, rather than static, mechanisms to actively regulate its level of stability. We propose that two specific adaptation mechanisms, spike frequency adaptation (SFA) and short-term synaptic depression (STD), continuously modulate the effective connectivity, keeping the brain near the edge of chaos and reducing dynamical fluctuations caused by stimuli. This theoretical framework links intrinsic and synaptic negative feedback mechanisms to network-level dynamics. This offers an explanation of why data-driven modeling of human brain signals, an exciting and useful method in epilepsy and anesthesiology research, seems to require linear time-varying (LTV) models which are refit every half second: difficult to observe adaptation processes interact with nonlinearities to make connectivity effectively dynamic at the macroelectrode scale. We suggest that compromised adaptation may underlie neurological conditions characterized by altered excitability, and that targeted brain stimulation could be used to probe the regulatory action of adaptation.