Table 4 The comparative pruning results for VGG-16, where the Acc.\(\downarrow\) indicates the percentage decrease in prediction accuracy after pruning. The best pruning results are bolded.
From: A multi-agent reinforcement learning based approach for automatic filter pruning
Datasets | Methods | Acc. (base) | Acc. (new) | Acc.\(\downarrow\) (%) |
|---|---|---|---|---|
CIFAR-10 | Random | 0.9356 | 0.8882 | 5.07 |
HRank | 0.9396 | 0.9234 | 1.72 | |
HBFP | 0.9396 | 0.9199 | 2.10 | |
DDPG_FP | 0.9367 | 0.9324 | 0.46 | |
QMIX_FP | 0.9356 | 0.9319 | 0.40 | |
CIFAR-100 | Random | 0.8961 | 0.8047 | 10.20 |
DDPG_FP | 0.7171 | 0.6912 | 3.61 | |
QMIX_FP | 0.8961 | 0.8876 | 0.95 |
