aiqbench analysis
GraViti introduces a transformer-based graph-level variational autoencoder that achieves state-of-the-art reconstruction on molecular datasets by relaxing permutation invariance requirements in the loss function. This architectural shift eliminates the computationally expensive cubic-cost matching step, reducing complexity to quadratic while preserving generalization through an invariant encoder. The resulting latent space enables efficient property-guided search and controlled molecular editing, offering a practical advance for generative chemistry tasks.
GraViti: Graph-Level Variational Autoencoders with Relaxed Permutation Invariance
arXiv:2605.16668v2 Announce Type: replace-cross Abstract: We introduce GraViti, a transformer-based graph-level variational autoencoder that encodes entire graphs into single, fixed-dimensional latent vectors rather than per-node embeddings, yielding a graph-level latent space that supports smooth interpolation, property-guided search, and other downstream tasks beyond the reach of node-level approaches. GraViti achieves state-of-the-art reconstruction accuracy on large molecular graph datasets while offering competitive single-step generative performance. Our central finding concerns where permutation invariance is needed in such a model, and where it is not. The encoder remains permutation-invariant throughout, ensuring the model generalizes to graphs regardless of input order. The reconstruction loss, however, does not need to be: when training data is provided in a consistent node order, comparing predictions to targets directly in that order, without a graph-matching step, is sufficient. Relaxing invariance in the loss alone lets GraViti avoid the cubic-cost matching step required by prior graph-level autoencoders, reducing complexity to quadratic while improving reconstruction fidelity. We show the resulting latent space is chemically meaningful, supporting controlled molecular editing and property optimization, recovering established chemical trends, and enabling direct regression of graph-level properties from latent embeddings.
Abstract:We introduce GraViti, a transformer-based graph-level variational autoencoder that encodes entire graphs into single, fixed-dimensional latent vectors rather than per-node embeddings, yielding a graph-level latent space that supports smooth interpolation, property-guided search, and other downstream tasks beyond the reach of node-level approaches. GraViti achieves state-of-the-art reconstruction accuracy on large molecular graph datasets while offering competitive single-step generative performance. Our central finding concerns where permutation invariance is needed in such a model, and where it is not. The encoder remains permutation-invariant throughout, ensuring the model generalizes to graphs regardless of input order. The reconstruction loss, however, does not need to be: when training data is provided in a consistent node order, comparing predictions to targets directly in that order, without a graph-matching step, is sufficient. Relaxing invariance in the loss alone lets GraViti avoid the cubic-cost matching step required by prior graph-level autoencoders, reducing complexity to quadratic while improving reconstruction fidelity. We show the resulting latent space is chemically meaningful, supporting controlled molecular editing and property optimization, recovering established chemical trends, and enabling direct regression of graph-level properties from latent embeddings.
Machine Learning (cs.LG); Artificial Intelligence (cs.AI)
arXiv:2605.16668 [cs.LG]
(or arXiv:2605.16668v2 [cs.LG] for this version)
https://doi.org/10.48550/arXiv.2605.16668 Focus to learn more arXiv-issued DOI via DataCite
arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.
Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.
Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs.
Full text under the publisher's open licence · 275 words
Related coverage
J++ Lens: Jacobian Filtering Enables More Faithful Workspace Lenses
LessWrong · Oct 8, 2026
Does Syntax Matter? A Graph-Augmented Variational Topic Model for Computational Social Sciences
arXiv · Sep 9, 2026
Solving the traveling salesman problem on $\lceil \log_2 (N-1)! \rceil$ qubits with a variational quantum Kolmogorov--Arnold network: exact-gradient CVaR training and generalization over time-dependent graphs
arXiv · Oct 15, 2026
Tropical Algebraic Geometry for Neuronal Representations: An Arakelov-Green Measure Based Descriptor for Graph Learning
arXiv · Aug 6, 2026