Table 3.
This table shows the accuracy of N-MNIST on several state-of-the-art algorithms.
| Method | Accuracy (%) |
|---|---|
| Lee et al.: Training SNN using backpropagation (Lee et al., 2016) | 98.74 |
| HATS (Sironi et al., 2018) | 99.1 |
| Active perception with DVS (Yousefzadeh et al., 2018) | 98.8 |
| Spatiotemporal backpropagation (Wu et al., 2018) | 98.78 |
| SLAYER (Shreshtha and Orchard, 2018a) | 99.2 |
| DECOLLE (Kaiser et al., 2020) | 96 |
| HM2-BP (Jin et al., 2019) | 98.84 |
| Spike based supervised gradient descent (Lee et al., 2020) | 99.09 |
| LISNN (Cheng et al., 2020) | 99.45 |
| Segmented probability-maximization (Liu et al., 2020) | 96.3 |
| Graph based object classification (Bi et al., 2019) | 99.0 |
| Learnable membrane time constants (Fang et al., 2020) | 99.61 |
| Collapsed images with ANN | 99.23 |
Clearly our method is among the state of the art.