Following the architecture presented in the paper, the autoencoder will expand the number of dimensions and then create a bottleneck which will reduce the dimensions to 10 (a common practice with autoencoders, see here) This architecture is a bit exaggerated for the task — you can use far less neurons for each layer A Sparse Autoencoder is a type of autoencoder that employs sparsity to achieve an information bottleneck. You are currently offline. Well, the denoising autoencoder was proposed in 2008, 4 years before the dropout paper (Hinton, et al. A non-negative and online version of the PCA was intro- duced recently [5]. Note that p