[CNN] tutorials - dsindex/blog GitHub Wiki

references
- intuitive explanation of CNN
- CNN tutorial in Korean
- math behind CNN
- develop ngram multichannel CNN for sentiment analysis
- Convolutional Neural Networks for Sentence Classification
- what do you mean by conv1d, conv2d, conv3d in CNN
- why does each filter learn different features in CNN
- keras.js
- depth-wise separable convolution
  - more clear explaination
  - depth-wise(channel-wise) and point-wise convolution for parameters reduction
    - depth-wise convolution : [height, width, in_channels]
    - point-wise(1x1) convolution : [1, 1, in_channels, out_channels]
    - ex) parameters : 3x3x16x32 -> depth-wise(3x3x16) + point-wise(1x1x16x32)
    - keras.layers.separable_conv2d
    - example codes
      - use case, tf.nn.separable_conv2d()
      - use case, SeparableConv2D()
    - for 2D input(ex, text processing), LxD -> Lx1xD expand -> depth-wise -> point-wise
implementations
- Implementing a CNN for Text Classification in TensorFlow
- Implementing a sentence classification using Char level CNN & RNN