基于强化学习和3σ准则的组合剪枝方法
|
|
徐少铭,李钰,袁晴龙
|
Combination pruning method based on reinforcement learning and 3σ criterion
|
|
Shao-ming XU,Yu LI,Qing-long YUAN
|
|
| 表 1 VGG16的各卷积层剪枝结果 |
| Tab.1 Every convolutional layer pruning result of VGG16 |
|
| 卷积层 | pt/kB | ac/% | | conv1-1 | 6.752 | 2.0 | | conv1-2 | 144 | 3.0 | | conv2-1 | 288 | 10.0 | | conv2-2 | 576 | 15.0 | | conv3-1 | 1 152 | 38.0 | | conv3-2 | 2 304 | 56.0 | | conv3-3 | 2 304 | 80.0 | | conv4-1 | 4 608 | 80.0 | | conv4-2 | 9 216 | 80.0 | | conv4-3 | 9 216 | 80.0 | | conv5-1 | 9 216 | 80.0 | | conv5-2 | 9 216 | 80.0 | | conv5-3 | 9 216 | 80.0 | | 总计 | 57 479.25 | 76.4 |
|
|
|