Skipping the Zeros in Diffusion Models for Sparse Data Generation
RPTU University Kaiserslautern-Landau · RPTU Kaiserslautern-Landau · Universidade de São Paulo · Ruprecht-Karls-Universität Heidelberg · University of California, Irivine · RPTU, Kaiserslautern · University of Kaiserslautern
PDF 由论文原始站点提供，PaperCompass 不保存论文文件。
摘要
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 training and inference. Across physics and biology benchmarks, SED matches or surpasses conventional DMs and domain-specific baselines, while vision experiments provide intuitive insights into the limitations of dense DMs and the benefits of SED.