This paper was accepted on the Diffusion Fashions workshop at NeurIPS 2023.
Rating-based fashions have shortly grow to be the de facto selection for generative modeling of pictures, textual content and extra just lately molecules. Nonetheless, to adapt a score-based generative modeling to those domains the rating community must be rigorously designed, hampering its applicability to arbitrary information domains. On this paper we deal with this drawback by taking a textit{purposeful} view of information. This purposeful view permits to solid seemingly completely different domains to a typical shared illustration. We then re-formulate the rating operate to cope with purposeful information and present: i) this unified structure might be successfully utilized to completely different modalities: pictures, geometry, video, and ii) we are able to study generative fashions of alerts outlined on non-euclidean geometry.