Energy-Based Cross Attention for Bayesian Context Update in Text-to-Image Diffusion Models Geon Y eong Park 1 Jeongsol Kim 1 Beomsu Kim 2 Sang Wan Lee 1,2,3

Nov-20-2025, 00:33:05 GMT–Neural Information Processing Systems 

Since diffusion models require the iterative sampling on high dimensional space, they are computationally expansive and time consuming.