What's the difference between convolutional neural network and convolutional auto encoders? What's the difference between convolutional neural network and convolutional auto-encoders?
This paper explains both but I find it hard to pinpoint the difference: http://people.idsia.ch/~juergen/icann2011stack.pdf
Many thanks!
 A: A convolutional autoencoder is a type of convolutional neural network. However people probably refer to convolutional neural networks as supervised convolutional neural networks, i.e. with a softmax classifier or a SVM classifier on top while a convolutional autoencoder is an unsupervised network, so it tries to reconstruct the input using learned features and might have an $L2$ error loss like $\|f(x) - x \|_2$ where $f(x)$ is the network output and $x$ is the input image. The term autoencoder is used to represent the fact that we are trying to reconstruct the input, and thus requires no labels. 
We design the conv autoencoder so that it does not simply learn the identity function (as in simply returning the input image without really producing any meaningful intermediate feature representation), for example sometimes by producing intermediate representations of a reduced dimensionality, among others methods.
A: Auto-encoders are models that learn the non-trivial identity function. What it means is that they try to learn a manifold on which the data lies on and can be used to generate the samples from the learned manifold. Convolution auto-encoders carry the same features except the fact that convolutional layers are present in the model. 
