Graph Generation with K²-trees

Overview

We introduce a graph generation method using K²-tree representation for compact graph synthesis. Our approach includes a sequential K²-tree representation that incorporates pruning, flattening, and tokenization processes, along with a Transformer architecture featuring specialized tree positional encoding. The method is evaluated on both general and molecular graph datasets, demonstrating effective and scalable graph generation.

BibTeX

@inproceedings{
  jang2024graph,
  title={Graph Generation with  \$K{\textasciicircum}2\$-trees},
  author={Yunhui Jang and Dongwoo Kim and Sungsoo Ahn},
  booktitle={The Twelfth International Conference on Learning Representations},
  year={2024},
  url={https://openreview.net/forum?id=RIEW6M9YoV}
}