Convolutional Neural Networks

Convolutional Neural Networks

1. ... is used to reduce amount of parameters of neural networks

2. ... are using filter (or kernel) to extract specific features of the input information

3. 1 0 -1 1 0 -1 1 0 -1

   • Vertical edge detection

4. 1 1 1 0 0 0 -1 -1 -1

   • horitzontal edge detection

5. Padding ...

   • padding adds a bit more importance into values in the edge or input matrix, by extending matrix dimension by zeros
   • Valid
     • no padding
   • Same
     • Pad so that input matrix is the same as output matrix

6. Output matrix size formula:

7. n - size of input matrix, f - size of filter matrix, p - size of padding ma, s - size of striding ma, n_c - number of channels in image, n_c' - number of filters used

8. Strided ...

9. Pooling layer

   • max pooling
   • average pooling
   • a layer in neural network, which replaces input matrix with operation (max or avg) as one single number
   • quick way to reduce parameters, as there are no parameters to learn in pooling

10. Why Convolution?

    • Parameter Sharing: A feature detector (such as a vertical edge detector) that's useful in one part of the image is probably useful in another part of the image
    • Sparsity of connections: In each layer, each output value depends only on a small number of inputs

Images and files in the map