A new arXiv preprint introduces SNACK, a sequential notation framework for probabilistic graph generation. The authors propose a string representation that allows sequence models to generate and model graphs carrying discrete node and edge features.

SNACK works by linearizing graphs: it encodes the nonzero entries of a graph into a sequence. This design is intended to make graph generation compatible with sequence-based probabilistic models.

The source is the paper's abstract, so the write-up does not include experimental results or comparisons with other graph-generation methods. The claims are limited to the proposed representation and its intended use.