The manuscript studies the role of feedback connections in pattern retrieval networks. This is done in a hierarchical Hopfield-type network. Then, different types of top-down feedback are investigated. The study addresses an important question in computational neuroscience. All reviewers found the results interesting. Besides its relevance to modelling of biology, it is also of potential interest for technical applications in hierarchical information retrieval. Two weaknesses of the manuscript is that it is only applied to relatively simple datasets, and the roles of the feedback components (push-pull) is not entirely clear. Nevertheless, due to the interesting approach, we decided that the manuscript is of potential interest for the NeurIPS audience, in particular for the biology-oriented community.