1. Revision: Diabetes dataset
2. Design 10 outputs using Sigmoid?
2.1 Output a Distribution of prediction with Softmax
2.2 Softmax Layer
Example,
2.3 Loss Function-Cross Entropy
Cross Entropy in Numpy
Cross Entropy in PyTorch
注意交叉熵损失,最后一层不要做激活
Mini-Batch: batch_size=3
4. MNIST Dataset
Implementation of classifier to MNIST Dataset
0. import package
1. prepare
通道
2. Design model
3. Construct Loss and Optimizer
4. Train and Test
有极限,对图像来说,全连接 对局部不太好。
自动特征提取CNN