- Vein structure
1. 结构
2. 损失
3. 优化
Question:
Why softmax
existing problems softmax
cross entropy loss of meaning
- Digression: After a clear principle, it can be improved
使用 softmax 的时候,有一个默认的前提,就是每个物体只属于一类
比如人脸识别,每张脸只能对应一个人
Thinking:
- sigmod binary classification of Loss, the difference between multi-classification Loss of softmax
- The difference sigmod and softmax
Euclidean loss
- Loss weighting function
- Sample weights
- To improve softmax loss