Research Article
Fully Kernected Neural Networks
Table 3
Validation accuracy of CNN, KNN, and our FKNN on CIFAR100.
| | Methods | Network-depth | Validation accuracy (%) | Improvement on CNN |
| | CNN | ResNet-18 | 71.92 | 0 | | CNN | ResNet-110 | 72.78 | 0 | | KNN | ResNet-110 | 77.51 | 4.73% | | FKNN (ours) | ResNet-18 | 79.59 | 7.67% |
| | CNN | DenseNet-40 | 75.58 | 0 | | KNN | DenseNet-40 | 75.08 | ā0.5% | | FKNN (ours) | DenseNet-40 | 79.17 | 3.59% |
| | CNN | GoogLeNet-22 | 73.35 | 0 | | KNN | GoogLeNet-22 | 79.16 | 5.81% | | FKNN (ours) | GoogLeNet-22 | 79.27 | 5.92% |
|
|
We mark the best results in bold values.
|
