Autoencoder Feature Selector basic example

This code implements the method described in "Autoencoder Inspired Unsupervised Feature" (Han 2018)

This code implements the method described in "Autoencoder Inspired Unsupervised Feature" (Han 2018). Four 3x3 pixel images are generated, then an autoencoder is trained with Row-Sparse Regularization on the encoder and Sparsity Regularization. The AE is tested by attempting to denoise noisy images. It can be seen that regularization provides smaller weights and biases for the network, but at the cost of a worse reconstruction.