AI RESEARCH
Skipping the Zeros in Diffusion Models for Sparse Data Generation
arXiv CS.LG
•
ArXi:2605.01817v1 Announce Type: new Diffusion models (DMs) excel on dense continuous data, but are not designed for sparse continuous data. They do not model exact zeros that represent the deliberate absence of a signal. As a result, they erase sparsity patterns and perform unnecessary computation on mostly zero entries. With Sparsity-Exploiting Diffusion (SED), we model only non-zero values, preserving sparsity. SED delivers computational savings while maintaining or improving generation quality by skipping zeros during.