Review for NeurIPS paper: Self-Supervised Few-Shot Learning on Point Clouds

Additional Feedback: l.56 'non-parametric' - as opposed to which parametric representations? Why is the absence of parameters important? E.g. is the spheres split along specific axis? What is d in R d and which value did you use in your experiments? Is it the number of integer class labels / number of quadrants per sphere always set to 4 as suggested in l.142?