Table 4 Task- and model-specific hyperparameters
From: Advancing spatio-temporal processing through adaptation in spiking neural networks
lr | # neurons | # layers | q | α (SLAYER) | c (SLAYER) | Ep. | [\({\tau }_{u}^{\,{\mbox{min}}},{\tau }_{u}^{{\mbox{max}}\,}\)] | [\({\tau }_{w}^{\,{\mbox{min}}},{\tau }_{w}^{{\mbox{max}}\,}\)] | dropout | τout | batch size | ||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
SHD | SE | 0.01 | 128/360 | 1/2 | 120 | 5 | 0.4 | 300 | [5, 25] | [60, 300] | 15% | 15 | 256 |
EF | 0.01 | 360 | 2 | 60 | 5 | 0.4 | 300 | [5, 25] | [60, 300] | 15% | 15 | 256 | |
LIF | 0.01 | 360 | 2 | – | 5 | 0.1 | 300 | [5, 150] | – | 15% | 15 | 256 | |
SSC | SE | 0.006 | 720 | 2 | 120 | 5 | 0.4 | 40 | [5, 25] | [60, 300] | 15% | 15 | 256 |
EF | 0.006 | 720 | 2 | 60 | 5 | 0.4 | 40 | [5, 25] | [60, 300] | 15% | 15 | 256 | |
LIF | 0.006 | 720 | 2 | – | 5 | 0.1 | 40 | [5, 150] | – | 15% | 15 | 256 | |
ECG | SE | 0.01 | 36 | 1/2 | 120 | 5 | 0.2 | 400 | [5, 25] | [60, 300] | 15% | 3 | 64 |
EF | 0.01 | 36 | 1 | 60 | 5 | 0.2 | 400 | [5, 25] | [60, 300] | 15% | 3 | 64 | |
LIF | 0.01 | 36 | 1 | – | 5 | 0.1 | 400 | [5, 150] | - | 15% | 3 | 64 | |
BSD | SE | 0.01 | 512 | 1 | 120 | 5 | 0.4 | 400 | [5, 25] | [60, 300] | 0% | 15 | 128 |
LIF | 0.006 | 510 | 1 | – | 5 | 0.2 | 400 | [5, 50] | - | 0% | 15 | 128 | |
spring-mass | SE/EF | 0.01 | [25, 3200] | 1 | 228/65 | 5 | 0.4 | 200 | [5, 25] | [60, 300] | 0% | [1, 20] | 256 |
LIF | 0.01 | [27, 3202] | 1 | – | 10 | 0.5 | 200 | [1, 25] | – | 0% | [1, 20] | 256 | |
LSTM | 0.001 | [13, 1600] | 1 | – | – | – | 200 | – | – | 0% | [1, 20] | 256 | |
audio comp. | SE/EF | 5 ⋅ 10−4 | 300 | 4 | 120/20 | 5 | 0.4 | 10 | [5, 25] | [30, 300] | 0% | [1, 10] | 128 |
LIF | 5 ⋅ 10−4 | 302 | 4 | – | 5 | 0.1 | 10 | [5, 100] | − | 0% | [1, 10] | 128 |
