Neuronal population bursts, in which large numbers of neurons fire nearly simultaneously, are increasingly recognized as playing important roles in sensory processing and memory formation. In many sensory systems, information is encoded not only by which neurons are active, but also by the temporal order in which distributed neural populations are recruited. How such temporally structured inputs are learned and represented at the population level remains unclear. Here, we study a recurrent spiking neural network model in which extended sequences of population bursts, termed superbursts, emerge as a collective dynamical property. Through Pavlovian conditioning with brief sequential activations of neuronal subpopulations, combined with neuromodulator-gated spike-timing-dependent plasticity, an initially random network self-organizes into an intricate hybrid architecture containing a strong feedforward backbone embedded within sparse recurrent connectivity. After conditioning, learned spatiotemporal inputs reliably evoke stimulus-specific superbursts, whereas temporally or spatially perturbed inputs produce weaker and qualitatively distinct responses. These superbursts provide an amplified and noise-tolerant population representation capable of supporting millisecond-scale temporal discrimination over a broad ($\sim 100\mathrm{ms}$) temporal range.

Close

PrevNext

View figure in article

• 
• 
• 
• 
• 
• 
• 
• 
• 
• 

• Biological neural networks
• Learning
• Memory
• Neuronal dynamics
• Neuronal network activity