Appendix: Energy-Based Cross Attention for Bayesian Context 388 Update in T ext-to-Image Diffusion Models 389 A Proof of Theorem 1 390 Theorem 1. For the energy functions 391 E(Q; K) = α 2 diag(KK

Oct-9-2025, 11:29:06 GMT–Neural Information Processing Systems 

During the reverse diffusion process in Sections C.2 and C.3, we kept