Table 3.
Overall results of proposed method and comparison methods. The bold numbers mean the best results.
| Model | Method | Error % | # W | Pruned W % | Memory (M) | CR |
|---|---|---|---|---|---|---|
| LeNet-300-100 | original | 2.26 | 266,610 | - | 3.08 | - |
| Naivecut | 2.84 | 106,937 | 59.89 | 1.46 | 2.49 | |
| Iterative pruning | 2.37 | 26,874 | 89.92 | 0.34 | 9.92 | |
| MONNP | 2.2 | 44,364 | 83.36 | 0.57 | 6.01 | |
| DENNC | 2.07 | 10,962 | 95.89 | 0.14 | 24.33 | |
| LeNet-5 | original | 0.95 | 45,278 | - | 0.55 | - |
| Naivecut | 1.71 | 19,198 | 57.5 | 0.26 | 2.35 | |
| Iterative pruning | 1.72 | 3826 | 91.55 | 0.05 | 11.83 | |
| MONNP | 0.91 | 8503 | 81.22 | 0.11 | 5.32 | |
| DENNC | 1.41 | 3444 | 93.09 | 0.04 | 14.47 | |
| AlexNet | original | 9.96 | 5,488,106 | - | 41.92 | - |
| Naivecut | 11.66 | 1,153,600 | 78.98 | 9.57 | 4.76 | |
| Iterative pruning | 11.51 | 275,503 | 94.98 | 2.19 | 19.92 | |
| MONNP | 11.41 | 748,578 | 86.36 | 5.99 | 7.33 | |
| DENNC | 11.59 | 192120 | 96.57 | 1.47 | 29.15 | |
| VGG16 | original | 12.43 | 2,112,730 | - | 37.29 | - |
| Naivecut | 12.94 | 972,278 | 53.98 | 18.83 | 2.17 | |
| Iterative pruning | 16.79 | 211,696 | 89.98 | 3.94 | 9.98 | |
| MONNP | 16.42 | 312,895 | 85.19 | 5.93 | 6.75 | |
| DENNC | 13.16 | 168,489 | 92.03 | 3.08 | 12.55 |