ProtoGuide: Prototype-Driven Guidance for Class-Conditional Graph Generation
Abstract
Discrete diffusion models are a prominent family for graph generation, but standard class-conditional mechanisms embed the class signal in the denoiser during training, tying the conditioning mechanism to the trained model. Classifier guidance avoids this coupling in continuous domains by steering a frozen model with a classifier's gradient, but discrete graph diffusion samples discrete edge states, so gradients cannot propagate through the sampled graph. We introduce ProtoGuide, a post-hoc, backbone-agnostic framework that recovers an analogous mechanism. At each reverse step the denoiser's per-edge output is relaxed into a differentiable soft adjacency, embedded by a frozen Siamese graph neural network, and scored against a target-class prototype and its nearest competitor; the resulting per-edge gradient, damped by a cosine schedule, is injected back into the denoiser output. All components stay frozen, so guidance is retargeted by supplying a different prototype. On five classes of real-world networks and two architecturally different backbones, EDGE and DiGress, ProtoGuide raises macro classification accuracy from 50.7% to 73.5% and from 73.6% to 83.8%, and outperforms DiGress's built-in conditional training under our configuration. Gains are largest where the unguided models are weakest, and are not uniform across classes. Per-graph coverage remains high in most settings, while distributional effects are class-dependent. A Best-of-N selection baseline matches this accuracy given enough oversampling, but at a substantial cost in graph diversity. An independently initialized classifier, a directionality test, and a few-shot analysis support target-directed steering and robustness to very small support sets.
Related
- Conditional Flow Matching for Generation of 3D Multi-variable Instantaneous Urban Microclimate Fields
- Framework-Conditional Open Materials Generation
- Nonclassical cutoff fluctuations in squeezed-light-driven high-harmonic generation
- Quantum Optics of harmonic generation in the strongly driven Jaynes-Cummings-type system
- High-Repetition-Rate Laser-Driven Fast Neutron Generation from Planar Liquid Heavy Water Jets
BibTeX
@misc{romano2026protoguide,
title = {{ProtoGuide: Prototype-Driven Guidance for Class-Conditional Graph Generation}},
author = {Salvatore Romano and Marco Grassia and Pietro Liò and Giuseppe Mangioni},
howpublished = {arXiv},
year = {2026},
month = sep,
doi = {10.48550/arxiv.2609.15239},
eprint = {2609.15239},
archivePrefix = {arXiv},
url = {https://arxiv.org/abs/2609.15239},
note = {CC BY 4.0},
}
RIS
TY - UNPB
TI - ProtoGuide: Prototype-Driven Guidance for Class-Conditional Graph Generation
AU - Salvatore Romano
AU - Marco Grassia
AU - Pietro Liò
AU - Giuseppe Mangioni
PY - 2026
DA - 2026/09/14
PB - arXiv
DO - 10.48550/arxiv.2609.15239
UR - https://arxiv.org/abs/2609.15239
AB - Discrete diffusion models are a prominent family for graph generation, but standard class-conditional mechanisms embed the class signal in the denoiser during training, tying the conditioning mechanism to the trained model. Classifier guidance avoids this coupling in continuous domains by steering a frozen model with a classifier's gradient, but discrete graph diffusion samples discrete edge states, so gradients cannot propagate through the sampled graph. We introduce ProtoGuide, a post-hoc, backbone-agnostic framework that recovers an analogous mechanism. At each reverse step the denoiser's per-edge output is relaxed into a differentiable soft adjacency, embedded by a frozen Siamese graph neural network, and scored against a target-class prototype and its nearest competitor; the resulting per-edge gradient, damped by a cosine schedule, is injected back into the denoiser output. All components stay frozen, so guidance is retargeted by supplying a different prototype. On five classes of real-world networks and two architecturally different backbones, EDGE and DiGress, ProtoGuide raises macro classification accuracy from 50.7% to 73.5% and from 73.6% to 83.8%, and outperforms DiGress's built-in conditional training under our configuration. Gains are largest where the unguided models are weakest, and are not uniform across classes. Per-graph coverage remains high in most settings, while distributional effects are class-dependent. A Best-of-N selection baseline matches this accuracy given enough oversampling, but at a substantial cost in graph diversity. An independently initialized classifier, a directionality test, and a few-shot analysis support target-directed steering and robustness to very small support sets.
ER -