CS231n笔记Lec5-卷积神经网络

Convolutional Neural Networks for Visual Recognition

Posted by Pekary on September 14, 2017 , views

CS231n笔记 Lec5 卷积神经网络

  • 在做图像识别时，只需zero-mean，不需要做normalization，也不需要做其他复杂的预处理

  • loss为nan表示损失太大，可能是学习率太大

  • Weight Initialization

    • Initialization too small: Activations go to zero, gradients also zero, No learning

    • Initialization too big: Activations saturate (for tanh), Gradients zero, no learning

    • Initialization just right:
      Nice distribution of activations at all layers, Learning proceeds nicely