AT able of Notation Description k Number of latent dimensions in hidden layer of autoencoder m Number of dimensions of input data n Number of datapoints W

Table 1: Summary of notation used in this manuscript, ordered according to introduction in main text. This can be justified by the following Lemma, Lemma 1. The proof is a simple application of the chain rule and Taylor's theorem. Thus, we need only compute the second derivative of the regularization terms. We proceed to take derivatives.