An experiment using an AlexNet like Converlution Neural Network(CNN) for Cifar10 classify task. Three different models are trained for comparison:
- Model1 : AlexNet like structure without dropout layers.
- Model2 : AlexNet with 0.5 dropout rate.
- Model3 : AlexNet with batch normalization layers.
- Model4 : Shallow ResNet with batch normalization layers.
| Models | Model1 | Model2 | Model3 |
|---|---|---|---|
| Random Crop | √ | √ | √ |
| Random Flip | √ | √ | √ |
| Kernel Size | 3×3 | 3×3 | 3×3 |
| Kernel Stride | 1 | 1 | 1 |
| Pooling | Max pooling | Max pooling | Max pooling |
| Pooling Stride | 2 | 2 | 2 |
| Conv layers | 3 | 3 | 3 |
| Dropout | × | 0.5 | × |
| Batch Normalization | × | × | √ |
- [1]. Alex Krizhevsky, Ilya Sutskever, Geoffrey E. Hinton. "ImageNet Classification with Deep Convolutional Neural Networks."
- [2]. Ioffe, Sergey, and Christian Szegedy. "Batch normalization: Accelerating deep network training by reducing internal covariate shift."
- [3]. Karen Simonyan, Andrew Zisserman. "Very Deep Convolutional Networks for Large-Scale Image Recognition."
- [4]. He, Kaiming, et al. "Deep Residual Learning for Image Recognition."
React
Typescript
Django
Laravel
Facebook
Microsoft
Google
Alibaba
D3
Tencent