04_Hung-yi Lee_What to do if my network fails to train2
Carpe Tu Black Whistle

Classification as Regression

  • Regression
    image
  • Classification as regression
    image

Class as one-hot vector

避免不同类,被认为是相互临近关系。
比如:
Class 1 = 1, Class 2 = 2, Class 3 = 3。
这种情况,容易被认为 Class 1 与 Class 2更加临近,与Class 3 没那么相似

所以使用: one-hot encode
但是存在缺点,就是编码冗余较大

image

  • Regression

image

  • Classification

image

Softmax

Loss of Classification

image

Mean Square Error(MSE)
Cross-entropy