Review for NeurIPS paper: Regularized linear autoencoders recover the principal components, eventually