Please enable JavaScript.

Coggle requires JavaScript to display documents.

教育訓練 (tensorflow (autodiff (reuse), session (interactive), name scop…

- - - - stochastic，every instance
        
        無法收斂到最佳
        
        逐漸減少learning rate
        
        fast
      - batch
        
        速度很慢
        
        可到global opt.
        
        可能陷到local min.
      - min-batch
    - - Large
        
        Jump out local min.
        
        Converge to min.
  - - - binary
        
        sigmod
      - mutliclass
        
        softmax
    - - \[-log(p) if y=1\]
      - \[-log(1-p) if y=0\]
  - - - recall
      - precise
      - accurcy
- - - - drop-out
      - regularization
    - - more generally the problem that the distribution of each layer’s inputs changes during training, as the parameters of the previous layers change (which they call the Internal Covariate Shift problem).
        
        Internal Covariate
  - - - filter/kernel
        
        Pooling
        
        strid
        
        padding
      - stacking feature maps
        
        Note that all neurons located in the same row i and column j but in different feature maps are connected to the outputs of the exact same neurons in the previous layer.
        
        不同feature map 上的neurons ,是不同維度的
        但input 卻是相同 (共線性)
        channel 不能相加, 因為不同維度
      - arch