Table 5.
ResNet50 on CIFAR10.
Method | Pruning ratio (%) | Val. acc. (%) | |
---|---|---|---|
Para. | FLOPs | ||
AutoML for model compression (93.53%) [49] | 60 | — | +0.12 |
Prune Train (94.2%) [25] | — | 70 | −1.10 |
Automatic neural network compression (93.55%) [26] | 50 | — | −1.51 |
Pruning it Yourself (92.85%) [50] | 41.52 | 34.18 | −0.16 |
MS-KD 0.6 (95.36%) [51] | 84.01 | 84.10 | −0.24 |
Ours scheme 1 Iteration-6 (93.6%) | 83.97 | 86.30 | +0.40 |