Questions for "With a Little Help from My Friends: Nearest-Neighbor Contrastive Learning of Visual Representations"
-----------------------------------------------------------------------------------------------

Please send your answers to: marrakch@cs.uni-freiburg.de by 13:30 on 24.01.2022

Question 1. (~2-3 sentences) What are the major differences between MoCo and NNCLR?

Question 2. (~2-3 sentences) Why is a strong data augmentation required to get good performances with MoCo? Why does the performance of NNCLR not suffer that much with a very simple augmentation pipeline?

Question 3. (~2 sentences) The transfer performance varies depending on the target dataset and the performance is lower on fine-grained datasets. Explain why