Table 2 Pruning results of AlexNet on CIFAR-10. The best pruning results are bolded, and the worst results are underlined.
From: A multi-agent reinforcement learning based approach for automatic filter pruning
\(\textrm{Imp}(\cdot )\) | Pruning rate: 50% | |||||
|---|---|---|---|---|---|---|
Acc.(pruned) | Acc.(fine-tuned) | Filters | Params(M) | FLOPs(M) | ||
|  | Base | 0.8585 | – | 1152 | 2.714826 | 39.823296 |
\(\varvec{w}_{i,j}\) | Weight | 0.8516 | 0.8480[98.78%] | 574 | 0.871555 | 14.119357 |
Taylor | 0.8203 | 0.8373[97.53%] | 551 | 0.754758 | 14.956772 | |
\(\varvec{F}_{i,j}\) | IB | 0.7734 | 0.8413[98.00%] | 570 | 0.880309 | 18.481240 |
BN | 0.8438 | 0.8470[98.66%] | 560 | 0.827679 | 18.891221 | |
Gradient | 0.8438 | 0.8520[99.24%] | 575 | 0.885663 | 21.158055 | |
\(\varvec{A}_{i,j}\) | APoP | 0.8516 | 0.8489[98.88%] | 550 | 0.860180 | 20.911694 |
\(\textrm{Imp}(\cdot )\) | pruning rate: 60% | |||||
|---|---|---|---|---|---|---|
Acc.(pruned) | Acc.(fine-tuned) | Filters | Params(M) | FLOPs(M) | ||
|  | Base | 0.8585 | – | 1152 | 2.714826 | 39.823296 |
\(\varvec{w}_{i,j}\) | Weight | 0.8906 | 0.8472[98.68%] | 459 | 0.596856 | 15.469136 |
Taylor | 0.8047 | 0.8349[97.25%] | 446 | 0.647585 | 16.149933 | |
\(\varvec{F}_{i,j}\) | IB | 0.7969 | 0.8346[97.21%] | 454 | 0.615360 | 15.795638 |
BN | 0.7969 | 0.8452[98.45%] | 456 | 0.631417 | 16.397015 | |
Gradient | 0.7969 | 0.8353[97.30%] | 460 | 0.633874 | 13.138188 | |
\(\varvec{A}_{i,j}\) | APoP | 0.7656 | 0.8423[98.11%] | 455 | 0.658449 | 16.962633 |
\(\textrm{Imp}(\cdot )\) | pruning rate: 70% | |||||
|---|---|---|---|---|---|---|
Acc.(pruned) | Acc.(fine-tuned) | Filters | Params(M) | FLOPs(M) | ||
|  | Base | 0.8585 | – | 1152 | 2.714826 | 39.823296 |
\(\varvec{w}_{i,j}\) | Weight | 0.8047 | 0.8321[96.92%] | 345 | 0.414934 | 7.052608 |
Taylor | 0.7109 | 0.8171[95.18%] | 337 | 0.393662 | 6.878678 | |
\(\varvec{F}_{i,j}\) | IB | 0.6250 | 0.7799[90.84%] | 328 | 0.380263 | 7.279463 |
BN | 0.7891 | 0.8326[96.98%] | 340 | 0.449226 | 12.486408 | |
Gradient | 0.7812 | 0.8366[97.45%] | 336 | 0.471698 | 15.172924 | |
\(\varvec{A}_{i,j}\) | APoP | 0.6328 | 0.8204[95.56%] | 334 | 0.408761 | 9.024843 |
