Blogs 2026-01 Generalization and Representation Learning in Diffusion Models. Representation learning is a strong bias/capability that enables diffusion models to generate novel and realistic samples. ICLR '26.
