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