Reviews: CliqueCNN: Deep Unsupervised Exemplar Learning

The paper reads well and is technically sound, especially the well-defined optimization problem in Sec. The idea of consolidating transitivity relationships into batches is an interesting angle in similarity learning. Excluding Section 2.3, however, the method as a whole feels a bit ad-hoc. There doesn't seem to be any reason why the assignment to batches could not be performed on samples directly, rather than cliques (other than that it might speed things up). It would make for a more well-justified method to introduced it in terms of the raw individual samples.