Table 6.
Performance measures for all networks and all tasks.
| Tasks/circuits | Data-based | Amorphous | Small-world | DC | DC (no io) | Random dynamics | Static synapses |
|---|---|---|---|---|---|---|---|
| tcl1(t) (L23) | 0.87 | –0.305 | –0.17 | –0.106 | –0.235 | –0.117 | 0.013 |
| tcl2(t) (L23) | 0.947 | –0.175 | –0.171 | –0.097 | –0.168 | –0.44 | –0.067 |
| tcl1(t) (L5) | 0.693 | –0.115 | –0.145 | –0.106 | –0.049 | –0.407 | –0.069 |
| tcl2(t) (L5) | 0.968 | –0.084 | –0.117 | –0.065 | –0.191 | –0.469 | –0.159 |
| tcl1(t − Δt) (L23) | 0.389 | –0.224 | –0.024 | –0.061 | –0.131 | +0.021 | –0.056 |
| tcl2(t − Δt) (L23) | 0.352 | –0.158 | –0.02 | +0.071 | –0.081 | –0.185 | –0.091 |
| tcl1(t − Δt) (L5) | 0.418 | –0.217 | –0.073 | –0.117 | –0.101 | –0.136 | –0.258 |
| tcl2(t − Δt) (L5) | 0.49 | –0.175 | –0.119 | –0.007 | –0.187 | –0.259 | –0.196 |
| XOR (L23) | 0.304 | –0.097 | –0.145 | –0.053 | –0.114 | –0.174 | –0.061 |
| r1/r2 (L23) | 0.058 | –0.05 | +0.083 | +0.043 | +0.076 | +0.073 | –0.007 |
| (r1 − r2)2 (L23) | 0.1 | –0.056 | +0.026 | +0.028 | +0.012 | +0.006 | –0.013 |
| XOR (L5) | 0.383 | –0.132 | –0.206 | –0.102 | –0.198 | –0.206 | –0.095 |
| r1/r2 (L5) | 0.051 | –0.04 | +0.069 | +0.064 | +0.116 | +0.029 | –0.025 |
| (r1 − r2)2 (L5) | 0.11 | –0.052 | +0.026 | +0.017 | +0.014 | –0.02 | –0.067 |
| Memory | 0.43 | –0.183 | –0.07 | –0.001 | –0.147 | –0.17 | –0.135 |
| Nonlinear | 0.168 | –0.072 | –0.025 | –0.001 | –0.016 | –0.049 | –0.045 |
| Other | 0.938 | –0.162 | –0.144 | –0.083 | –0.196 | –0.373 | –0.093 |
| All | 0.463 | -0.129 | -0.072 | -0.024 | -0.105 | -0.176 | -0.084 |
Spike based tasks are evaluated with the kappa coefficient and rate based tasks with the correlation coefficient. The data-based column gives the absolute value and the other columns show the difference from this value. The best performance per task/row is marked in bold. Gray/blue shading denotes tasks from the categories memory/nonlinear. All values are averaged over 20 runs (10 runs in original paper).