Table 1 Comparison of recurrent LIF and adLIF networks on spike-encoded speech recognition datasets
From: Advancing spatio-temporal processing through adaptation in spiking neural networks
Model | Publication | #Params | Test Acc. [%] | |
|---|---|---|---|---|
SHD | LIF | Cramer et al. 202235 | N.A. | 83.2 ± 1.3 |
LIF | Deckers et al. 202424 | 37.9k | 84.49 | |
LIF | Bittar & Garner 202216 | 141k | 87.04 | |
LIF | Bittar & Garner 202216 | 3.8M | 89.29 | |
LIF | This work | 450k | 90.27 ± 0.73 | |
SE-adLIF | This work | 450k | 95.81 ± 0.56 | |
SSC | LIF | Cramer et al. 202235 | N.A. | 50.9 ± 1.1 |
LIF | Deckers et al. 202424 | 0.34M | 71.76 | |
LIF | Bittar & Garner 202216 | 141k | 66.67 | |
LIF | Bittar & Garner 202216 | 1.1M | 68.14 | |
LIF | This work | 1.7M | 75.23 | |
SE-adLIF | This work | 1.6M | 80.44 ± 0.26 | |
ECG | LIF | This work | 1.8k | 77.8 ± 1.85 |
SE-adLIF | This work | 1.8k | 86.88 ± 0.40 |
