Table 6 Performance comparison with different batch size.
From: Enhancing deep neural network training efficiency and performance through linear prediction
| Â | VGG16 | Resnet18 | GoogLenet | |||||||
|---|---|---|---|---|---|---|---|---|---|---|
Batch size | DEMON | PLP | SGD | DEMON | PLP | SGD | DEMON | PLP | SGD | |
32 | epoch | 34 | 35 | 33 | 30 | 27 | 28 | 33 | 32 | 35 |
Acc/% | 61.23 | 61.27 | 60.65 | 64.49 | 65.35 | 64.28 | 66.06 | 65.85 | 65.06 | |
top-1 | 0.3877 | 0.3873 | 0.3935 | 3.551 | 3.465 | 3.572 | 0.3394 | 0.3435 | 0.3494 | |
top-5 | 0.1265 | 0.1319 | 0.1328 | 1.184 | 1.078 | 1.073 | 0.1055 | 0.1125 | 0.1157 | |
64 | epoch | 35 | 35 | 35 | 36 | 37 | 40 | 38 | 35 | 38 |
Acc/% | 59.95 | 59.92 | 59.84 | 63.37 | 64.07 | 63.35 | 62.41 | 62.86 | 61.92 | |
top-1 | 0.4005 | 0.4009 | 0.4016 | 3.663 | 3.593 | 3.665 | 0.3759 | 0.3714 | 0.3787 | |
top-5 | 0.1383 | 0.1388 | 0.1389 | 1.116 | 1.152 | 1.234 | 0.1296 | 0.1203 | 0.1219 | |
128 | epoch | 34 | 35 | 37 | 46 | 46 | 48 | 40 | 40 | 44 |
Acc/% | 56.92 | 57.10 | 56.16 | 61.24 | 61.67 | 60.21 | 58.57 | 59.54 | 58.38 | |
top-1 | 0.4308 | 0.4290 | 0.4384 | 0.3976 | 0.3833 | 0.3979 | 0.4143 | 0.4046 | 0.4162 | |
top-5 | 0.1602 | 0.1600 | 0.1657 | 0.1326 | 0.1326 | 0.1316 | 0.1518 | 0.1416 | 0.1554 | |
256 | epoch | 37 | 35 | 39 | 62 | 59 | 65 | 52 | 55 | 54 |
Acc/% | 53.43 | 53.47 | 53.13 | 59.75 | 58.93 | 57.84 | 53.40 | 54.58 | 53.65 | |
top-1 | 0.4657 | 0.4653 | 0.4687 | 4.035 | 4.194 | 4.157 | 0.4664 | 0.4542 | 0.4635 | |
top-5 | 0.1850 | 0.1860 | 0.1875 | 1.580 | 1.607 | 1.623 | 0.1942 | 0.1784 | 0.1845 | |
