Table 2 Comparison of our SSM model to other leading models on the DVS128 Gesture and DVS128 Lips datasets in simulation
From: Compute-in-memory implementation of state space models for event sequence processing
Method | DVS128 Gesture | DVS128 Lips | ||||
|---|---|---|---|---|---|---|
Test Acc. | Params | Async. | Test Acc. | Params | Async. | |
SNN-based methods | ||||||
She et al.65 | 98.0% | 1.1 M | ✗ | – | – | – |
Apolinario et al.66 | 97.7% | 1.6 M | ✗ | – | – | – |
Wang et al.67 | 97.1% | 1.5 M | ✗ | 42.2% | 8.1 M | ✗ |
Bulzomi et al.46 | – | – | – | 60.2% | 47.0 M | ✗ |
RNN-based methods | ||||||
Innocenti et al.68 | 97.7% | – | ✗ | – | – | – |
Subramoney et al.69 | 97.8% | 4.8 M | ✗ | – | – | – |
CNN-based methods | ||||||
Tsourounis et al.70 | – | – | – | 63.2% | 40.5 M | ✗ |
Wang and Zhao et al.71 | – | – | – | 34.5% | 64.6 M | ✗ |
Tan et al.4 | – | – | – | 72.1% | 38.5 M | ✗ |
Other methods | ||||||
Martin-Turrero et al.72 | 94.1% | 14 M | ✓ | – | – | – |
Peng et al.73 | 97.9% | 4.5 M | ✗ | 69.8% | – | ✗ |
Gehrig et al.47 | – | – | – | 48.7% | 21.5 M | ✗ |
Liu et al.74 | 98.8% | – | ✗ | – | – | – |
SSM-based methods | ||||||
Schöne et al.36 | 97.7% | 5 M | ✓ | – | – | – |
Ours | 97.3% | 5 M | ✓ | 63.5% | 5.7 M | ✓ |
