Three types of layers: - C: convolutional layer (matrix map) - MP: max-pooling layer (matrix map) - F: fully connected layer (vector map) - O: output layer
Convolutional Layers: - Scale: scale (size of patch) - Number of output maps: outputMap - Shared weights: k - Bias: b
Max-pooling layer - Scale: scale (size of patch) - Max-coordinate matrix: k (1 if max, 0 if not)
Fully connected layer (dimension and number of feature maps stay the same) - Weight matrix: w - Bias: b
Output layer (dimension equal to the dimension of output label) - Weight matrix: w - Bias: b
Common parameters - Result: a - Delta: d