Table 7.
Model description of the network benchmark using the reference synapse model. After Nordlie et al. (2009).
| A: MODEL SUMMARY | ||
| Populations | Three: uncorrelated input (U), correlated input (C), target (T) | |
| Topology | Feed-forward | |
| Connectivity | All-to-one | |
| Neuron model | Leaky integrate-and-fire, fixed voltage threshold, fixed absolute refractory period (voltage clamp) | |
| Synapse model | Exponential-shaped postsynaptic conductances | |
| Plasticity | Intermediate Gütig spike-timing dependent plasticity | |
| Input | Fixed-rate Poisson (for U) and multiple interaction process (for C) spike trains | |
| Measurements | Synaptic weights | |
| B: POPULATIONS | ||
| Name | Elements | Population size |
| U | Parrot neurons | Nu |
| C | Parrot neurons | Nc |
| T | IAF neurons | NT |
| C: CONNECTIVITY | ||
| Source | Target | Pattern |
| U | T | All-to-all, uniformly distributed initial weights w, STDP, delay d |
| D: NEURON AND SYNAPSE MODEL | ||
| Name | IAF neuron | |
| Type | Leaky integrate-and-fire, exponential-shaped synaptic conductances | |
| Sub-threshold dynamics | Cm dV/dt = gL (EL − V) + g(t) (Ee − V) if t > t* + τref V(t) = Vreset else g(t) = wgmax exp(−t/τsyn) | |
| Spiking | If V(t−) < þeta ∧ V(t+) ≥ θ 1. Set t* = t, 2. Emit spike with time stamp t* | |
| Name | Parrot neuron | |
| Type | Repeats input spikes with delay d | |
| E: PLASTICITY | ||
| Name | Intermediate Gütig STDP | |
| Spike pairing scheme | Reduced symmetric nearest-neighbor | |
| Weight dynamics | ||
| x(Δt) = exp(−|Δt|/τSTDP) | ||
| F(w) = λ(1 − w)μ if Δt > 0 | ||
| F(w) = −λαwμ if Δt < 0 | ||
| F: INPUT | ||
| Type | Target | Description |
| Poisson generators | U | Independent Poisson spike trains with firing rate ρ |
| MIP generators | C | Spike trains with correlation c and firing rate ρ |
| G: MEASUREMENTS | ||
| evolution and final distribution of all synaptic weights | ||
For details about the hardware-inspired synapse model see Section 2.6.1.