Balancing Symmetry and Efficiency in Graph Flow Matching

2026-02-20

AI Summary

This work addresses the limitations of strict equivariance in graph generation models, which, while preserving symmetry, incurs high computational costs, slow convergence, and overfitting. To mitigate these issues, the authors propose a controllable symmetry modulation mechanism that dynamically relaxes symmetry constraints during training by integrating sinusoidal positional encodings, node permutations, and a discrete flow matching framework. This approach accelerates early-stage convergence, effectively suppresses shortcut learning and repetitive generation, and significantly delays overfitting. Experimental results demonstrate that the method achieves superior performance using only 19% of the baseline’s training epochs, substantially improving both training efficiency and generation diversity.

Technology Category

• Machine Learning: Deep Generative Models & Autoencoders
• Computer Vision: Generative Adversarial Networks (GANs) for Vision
• Natural Language Processing: Generation

Application Category

• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• Economics, Online Markets and Human Computation: Economic ramifications for generative AI infrastructure and applications

Abstract

Equivariance is central to graph generative models, as it ensures the model respects the permutation symmetry of graphs. However, strict equivariance can increase computational cost due to added architectural constraints, and can slow down convergence because the model must be consistent across a large space of possible node permutations. We study this trade-off for graph generative models. Specifically, we start from an equivariant discrete flow-matching model, and relax its equivariance during training via a controllable symmetry modulation scheme based on sinusoidal positional encodings and node permutations. Experiments first show that symmetry-breaking can accelerate early training by providing an easier learning signal, but at the expense of encouraging shortcut solutions that can cause overfitting, where the model repeatedly generates graphs that are duplicates of the training set. On the contrary, properly modulating the symmetry signal can delay overfitting while accelerating convergence, allowing the model to reach stronger performance with $19\%$ of the baseline training epochs.

Problem

• equivariance
• graph generative models
• symmetry
• overfitting
• training efficiency

Innovation

• symmetry modulation
• equivariance relaxation
• graph flow matching
• positional encoding
• graph generative models