Review for NeurIPS paper: CrossTransformers: spatially-aware few-shot transfer

Summary and Contributions: Few-shot learning is a challenging problem, which requires performing a supervised learning task with a small labeled support set. The classic few-shot learning problem has been extended (so to say) for an episodic learning setting by Triantafillou et al. The proposed cross-transformer framework addresses this episodic few-shot learning problem. The approach builds on the Prototypical Nets (Snell et al). Specifically, it makes the following contributions Identification of supervision collapse.