Adapting Precomputed Features for Efficient Graph Condensation

Graph Neural Networks (GNNs) face significant computational challenges when handling large-scale graphs. To address this, Graph Condensation (GC) methods aim to compress large graphs into smaller, synthetic ones that are more manageable for GNN training. Recently, trajectory matching methods have shown state-of-the-art (SOTA) performance for GC, aligning the model's training behavior on a condensed graph with that on the original graph by guiding the trajectory of model parameters. However, these approaches require repetitive GNN retraining during condensation, making them computationally expensive. To address the efficiency issue, we completely bypass trajectory matching and propose a novel two-stage framework. The first stage, a precomputation stage, performs one-time message passing to extract structural and semantic information from the original graph. The second stage, a diversity-aware adaptation stage, performs class-wise alignment while maximizing the diversity of synthetic features. Remarkably, even with just the precomputation stage, which takes only seconds, our method either matches or surpasses 5 out of 9 baseline results. Extensive experiments show that our approach achieves comparable or better performance while being 96× to 2,455× faster than SOTA methods, making it more practical for large-scale GNN applications. Our code and data are available at https://github.com/Xtra-Computing/GCPA.

Training deep learning models on large-scale graphs—such as those used in social networks or e-commerce platforms—can be extremely time-consuming and computationally intensive. To improve efficiency, researchers have developed techniques that compress these large graphs into smaller versions while preserving their most important information. The most effective existing methods typically rely on iterative training, where the model is repeatedly optimized to mimic the behavior of the original graph. While this can yield high accuracy, the process is often slow and, for some large graphs, can require substantial computational time. Our method offers a significantly faster alternative: rather than retraining multiple times, we extract useful representations from the large graph in a single step and use them to build a condensed version. This approach achieves up to a 2,455× speedup over SOTA methods, while matching or even surpassing their accuracy. It provides a practical and scalable solution for efficient learning on large graphs.