Suppose, you are training a deep convolutional neural networ…
Suppose, you are training a deep convolutional neural network to classify hand-written characters (10 numerical digits + 26 letters in the alphabet). Your network has 12 feature extraction layers (a series of convolution and pooling layers). At the end of those layers a feature map of dimension 4 x 4 x 32 (Height x Width x No. of features) is generated. These features are fed to a single fully-connected layer to generate the output probabilities. What is the total number of parameters in the output layer of your network? Note, the input images are gray scale images with dimensions 20 pixels x 20 pixels. (show your calculation as detailed as possible in the box below- partial points available). Please use the illustration below shown in lectures as your reference. It does not pertain to the problem in hand, but illustrates an example network for MNIST digit classification
Read Details