Introduction Recurrent connectivity and nonlinearity make neural networks inherently susceptible to destabilization by fluctuating input, yet the brain must maintain a consistent level of stability. Near the edge of chaos, decodable information persists over extended timescales. However, in sparse networks obeying Dale's law, structural balance alone cannot constrain destabilizing eigenvalue outliers [1]. Furthermore, external stimuli can alter stability, especially in nonlinear networks [2]. We hypothesized that two complementary forms of adaptation, spike frequency adaptation (SFA) and short-term synaptic depression (STD), together regulate network stability.
Methods
Results Only networks with both SFA and STD consistently operated near the edge of chaos even as connectivity parameters varied and external stimulation changed. Networks without dual adaptation were much more likely to be overly stable or highly chaotic as connectivity parameters were varied. During excitatory stimulation, networks with no adaptation or with SFA only became significantly more chaotic [2]. As a consequence of remaining near the edge of chaos, networks with both SFA and STD had the greatest memory capacity.
Discussion SFA and STD provide complementary stabilizing mechanisms that together maintain near-edge-of-chaos dynamics and maximize memory capacity [2]. Multi-timescale SFA approximates fractional differentiation [3,4], connecting our framework to fractional-order dynamical models. Fractional-order network models distinguish epileptic brain states, and stabilizing their dynamics suppresses seizures in simulation [5]. EEG recordings near the seizure onset zone show power spectral density slope changes consistent with altered adaptation [6], suggesting that adaptation dysfunction may contribute to epilepsy.
References
Acknowledgement This work was supported by funding from the NIH awarded to B.N.L. (NINDS R01NS129622 and K23NS112339).
