an advantage of our approach on representation learning for downstream tasks which was considered difficult for AEs

We thank all reviewers for acknowledging the novelty and contributions of our work. As this is the first work of applying implicit regularization method, there could be many follow-up questions to explore. We will study this effect in our future work. Our model outperforms strong baselines such as W AE [1] and RAE [2]. AEs perform differently with varying latent dimension.