Reconstruction Entropy

After applying Gaussian noise, a kind of statistical white noise, to the data, this objective function punishes the network for any result that is not closer to the original input. That signal prompts the network to learn different features in an attempt to reconstruct the input better and minimize error.