1 paper
Alexander C. Li, Yuandong Tian, Beidi Chen +2
Conventional wisdom suggests that pre-training Vision Transformers (ViT) improves downstream performance by learning useful representations. Is this actually true? We investigate t…