The resting brain must preserve stable representations while flexibly switching among activity patterns. Here we show that noise can tune this balance in a connectome-informed modular spiking network storing multiple spatiotemporal memories through spike-timing-dependent plasticity. With global inhibition fixed, varying excitatory coupling and noise reveals a noise-controlled sequence of retrieval transitions: no effective retrieval, a first-order-like hysteretic transition, a fluctuation-dominated continuous-like regime with critical-like signatures, a reentrant first-order-like transition, and noise-dominated disorder. At intermediate noise, multiple stored patterns are transiently reactivated, producing recurrent spontaneous switching among memory states. Time-dependent noise modulation further drives the same learned network between persistent replay, disordered firing, and multipattern switching. The qualitative transition sequence persists after degree-preserving randomization of the structural connectome, indicating robustness to the detailed empirical topology. A reduced phenomenological model based on memory-overlap dynamics interprets the switching regime as arising from a balance between noise-assisted activation of memory attractors and noise-induced escape from them. These results identify noise as a control parameter for retrieval transitions, attractor stability, and flexible memory-state exploration in modular brain networks.

Close

PrevNext

View figure in article

• Dynamics of networks
• Network phase transitions
• Neuronal network activity