Table 1.
Performance as a function of the number of patterns P.
| P | 5 | 10 | 20 | 40 |
|---|---|---|---|---|
| Δtopt (ms) | 11 | 8.1 | 5.7 | 3.7 |
| τopt (ms) | 8.9 | 6.8 | 5.6 | 5.1 |
| Mopt | 1,600 | 2,300 | 3,100 | 3,800 |
| SNRopt | 31 | 20 | 12 | 6.7 |
| θ0 | 190 | 140 | 110 | 92 |
| wout | −6.210−3 | −6.310−3 | −6.510−3 | −6.710−3 |
| <Plearned> | 5 | 10 | 20 | 39.5 |
| Hit rate (%) | 98.9 | 98.6 | 97.9 | 96.5 |
| False alarms (Hz) | 0 | 0 | 0 | 0 |
| P(opt) (%) | 100 | 100 | 100 | 58 |
The first four lines are computed from the theoretical optimum. The next two lines are the optimal values found through exhaustive search (see text). The last four lines are performance indicators, estimated during the last 100 presentations of each pattern. 〉Plearned〈 is the mean number of “learned patterns,” that is by convention patterns which elicit at least one postsynaptic spike. The following line is the mean hit rate for those patterns. The subsequent line gives the false alarm rate, but we never observed any here. Finally P(opt) is the proportion of optimal cases.