CE / MSE

  1. why using CrossEntropy for classification problem, while Mean Square Error for regression problem?
    CE loss:
    ML - 图2
    where ML - 图3. For ML - 图4, if ML - 图5 is closer to 0, the ML - 图6 is getting smaller and the reducing speed is increasing. ML - 图7. That means that more distance between p and y is , more and more the loss will be.
    For MSE, the loss is correspond to the bias between p and y.

    LSTM

    Screenshot_20210224_170642.png
    ML - 图9

    GRU

    Screenshot_20210224_165633.png

    ML - 图11