1 paper · 1 filter
Zhiwei Tang, Jiangweizhi Peng, Jiasheng Tang +3
In this work, we focus on the alignment problem of diffusion models with a continuous reward function, which represents specific objectives for downstream tasks, such as increasing…