From: An efficient pruning scheme of deep neural networks for Internet of Things applications
Architecture | Method | Baseline (%) | Accuracy (%) | FLOPs pruned (%) | Parameters pruned (%) |
---|---|---|---|---|---|
ResNet-32 | SFP[37] | 92.63 | ↓ 0.55 | 41.5 | - |
 | FPGM[45] | 92.63 | ↓ 0.70 | 53.2 | - |
 | COP[33] | 92.64 | ↓ 0.67 | 53.9 | 57.5 |
 | Ours | 93.20 | ↓ 0.21 | 33.0 | 31.7 |
 |  |  | ↓ 0.70 | 49.0 | 60.1 |
ResNet-56 | PFEC[18] | 93.04 | ↑ 0.02 | 27.6 | 13.7 |
 | CFP[32] | 93.57 | ↓ 0.25 | 61.5 | - |
 | DCP[30] | 93.80 | ↓ 0.31 | 50.3 | 50.7 |
 | PFS[38] | 93.23 | ↓ 0.18 | 50.0 | - |
 | ASS[39] | 93.26 | ↓ 0.03 | 54.1 | 54.2 |
 | GAL[35] | 93.26 | ↓ 1.68 | 60.2 | 65.9 |
 | Ours | 93.65 | ↑ 0.08 | 35.0 | 41.1 |
 |  |  | ↓ 0.17 | 49.6 | 58.0 |
 |  |  | ↓ 0.51 | 56.7 | 62.2 |