GraViti: Graph-Level Variational Autoencoders with Relaxed Permutation Invariance

2026-05-15

AI Summary

This work proposes GraViti, a Transformer-based graph-level variational autoencoder that addresses key limitations of conventional graph generation models, which rely on node-level embeddings and enforce permutation invariance—often at the expense of reconstruction fidelity, particularly in domains with inherent node orderings such as molecular graphs. By relaxing the strict permutation invariance constraint and leveraging natural node sequences, GraViti enables higher-quality graph reconstruction through an end-to-end graph-to-vector mapping and a single-step decoding mechanism. The resulting latent space is compact, smooth, and amenable to manipulation, supporting property-guided generation and interpolation. Evaluated on large-scale molecular datasets, GraViti achieves state-of-the-art reconstruction accuracy while efficiently generating chemically valid molecules.

Abstract

We introduce GraViti, a transformer-based graph-level variational autoencoder that maps entire graphs to compact latent vectors. This design produces a true graph-level latent space that supports smooth interpolation, property-guided search, and other downstream tasks beyond the constraints of node-level embeddings. On molecular benchmarks, GraViti learns to decode valid samples that follow the chemical constraints present in the training data, showing that the model recovers domain rules directly from graph-level representations. We also show that, in domains where a reliable canonical node ordering exists such as molecules or bayesian networks, enforcing permutation invariance can prove detrimental for consistent reconstruction. GraViti achieves state-of-the-art reconstruction accuracy on large datasets, and provides solid generative performance. Its single-step decoding offers a lightweight alternative to more complex generation pipelines while maintaining practical sample quality.

Problem

• graph-level representation
• permutation invariance
• variational autoencoder
• graph generation
• molecular graphs

Innovation

• graph-level VAE
• relaxed permutation invariance
• transformer-based graph generation
• single-step decoding
• molecular graph representation