Review for NeurIPS paper: Continual Deep Learning by Functional Regularisation of Memorable Past

What are the real contributions of the paper? The idea of regularizing the outputs (or functional-regularization) has already been explored, as already said in the paper. Combining the idea of regularizing the outputs with memory-based methods is also already explored. Please see GEM [1] and A-GEM [2]. What makes this approach better or important, e.g.