Asynchronous simulation of large networks of spiking neurons and dynamical synapses

摘要

Numerical simulations of networks of `integrate and fire' (IF) neurons are typically `synchronous': for a given time resolution triangle opent the differential equations expressing the neural dynamics are numerically solved, with step triangle opent. In practical terms, the simplest integration methods can be used, and simulations of reasonable large networks (of order 10~4 neurons) are within the capabilities of a fast workstation. The need to introduce synaptic dynamics coupled to the evolution of neural states, makes large scale simulations unfeasible. There are two main reasons: 1) the number of synapses grows with the square of the number of neurons, at fixed connectivity. With synapses as dynamical variables, the computational load becomes excessive. 2) the typical time scale for synaptic modifications is generally assumed to be much longer than that of neural dynamics. Thus, to observe effects of synaptic dynamics on neural activities, much longer simulations (in `biological times') are required.