How do dynamic synapses shape neural network behavior? This research explores the role of frequency-dependent synaptic transmission, common in neocortical synapses, in neural network computation. The study proposes a unified phenomenological model that captures both fast depression and facilitation of synaptic transmission in response to varying action potential (AP) patterns. Using this model, the authors analyze different regimes of synaptic transmission, demonstrating that dynamic synapses transmit distinct aspects of presynaptic activity based on average frequency. The model also allows for deriving mean-field equations governing the activity of large, interconnected networks. The dynamics of synaptic transmission can result in a complex sets of regular and irregular regimes of network activity. This research provides a valuable framework for understanding how synaptic dynamics contribute to the rich computational capabilities of neural networks. The model and analyses presented offer insights for neuroscientists and artificial intelligence researchers alike, opening avenues for designing more biologically realistic and efficient artificial neural systems.
Published in Neural Computation, this research aligns with the journal's focus on computational and theoretical aspects of neural systems. By presenting a model of dynamic synapses and analyzing its impact on network activity, the study contributes to the journal's core themes of neural computation and information processing.