<ul><li>Neuro-Symbolic Diffusion (NSD) introduced as a framework to generate samples under physical, structural, and operational constraints.</li><li>Combines diffusion steps with symbolic optimization for generating consistent outputs adhering to user-defined constraints.</li><li>NSD supports continuous (e.g., images, trajectories) and discrete (e.g., molecular structures, natural language) output generation.</li><li>Demonstrated applications include safety, data scarcity, and out-of-domain generalization challenges.</li></ul>

Neuro-Symbolic Generative Diffusion Models for Physically Grounded, Robust, and Safe Generation

Discover more