1 paper
Zhipeng Yao, Dazhou Li, Zitong Zhang +6
Diffusion models have achieved remarkable success, yet their training remains inefficient due to a severe optimization bottleneck, which we term Representation Degradation. As nois…