A tiny MATLAB convolutional neural network
The network can be composed by the following layers:
- convolution
- pooling (max)
- normalization (relu)
- fully connected
Weights are learned using backpropagation
X_model.matcontains the weights learned on a training set of 20 examples and 300 backpropagation iterationsclassify.mimplements a 10-layers network trained to recognize crosses in 9x9 picturesdemo_class.mshows how to classify picturesdemo_train.mshows how to train the network via backpropagationplotCNN.mdisplays the layers that compose the network (see figure below)
