|
There is a lot of information available out there that talks about using CNNs for character classification tasks. In most of places a typical CNN architecture is used suggested by LeCun in "Gradient based learning applied to document recognition" Does anyone know about other applications where different versions (i.e. different # of feature maps per layer, and/or different # of convolution+max pooling layers) have been successfully used? I am looking for a source that tells me if there is any relations between a classification problem at hand and the choice of above mentioned architectural parameters of the CNN. |
