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 6 ResNet56全连接层不同阈值下的实验结果

Tab.6 Experimental results of different thresholds on ResNet56 fully connected layer

阈值 t af/% ao/% pp/% ∆p/% μ 0.197 1 56.86 1.30 68.28 −3.22 σ 0.148 9 45.70 1.05 68.66 −2.84 μ+σ 0.345 6 83.66 1.92 66.65 −4.85 μ+2σ 0.499 0 95.56 2.20 65.33 −5.76 μ+3σ 0.646 1 99.07 2.28 35.48 −35.61