Residual Connections Harm Generative Representation Learning 事件

Residual Connections Harm Generative Representation Learning arXiv:2404.10947v5 Announce Type: replace Abstract: We show that introducing a weighting factor to reduce the influence of identity shortcuts in residual networks significantly enhances semantic feature learning in generative representation learning frameworks, such as masked autoencoders (MAEs) and diffusion models. Our modification notably improves feature quality, raising ImageNet-1K K-Nearest Neighbor accuracy from 27.4% to 63.9%