Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Towards Faster Graph Partitioning via Pre-training and Inductive Inference

Meng Qin, Chaorui Zhang, Yu Gao, Yibin Ding, Weipeng Jiang, Weixi Zhang, Wei Han, Bo Bai

TL;DR

Graph partitioning is framed as a modularity-maximization problem and is typically NP-hard, motivating a scalable solution for large graphs. PR-GPT combines offline pre-training on small graphs with inductive online generalization to large graphs, followed by online refinement using fast GP refiners on a reduced, weighted super-graph, to accelerate partitioning without retraining. The approach demonstrates substantial efficiency gains on the IEEE HPEC Graph Challenge benchmark, often with negligible quality loss, and shows potential for streaming GP by progressively reducing the processing scale. The work contributes a scalable, transferable GP framework, aligns with foundation-model paradigms, and provides open-source tooling to enable efficient GP on massive graphs and streaming scenarios.

Abstract

Graph partitioning (GP) is a classic problem that divides the node set of a graph into densely-connected blocks. Following the IEEE HPEC Graph Challenge and recent advances in pre-training techniques (e.g., large-language models), we propose PR-GPT (Pre-trained & Refined Graph ParTitioning) based on a novel pre-training & refinement paradigm. We first conduct the offline pre-training of a deep graph learning (DGL) model on small synthetic graphs with various topology properties. By using the inductive inference of DGL, one can directly generalize the pre-trained model (with frozen model parameters) to large graphs and derive feasible GP results. We also use the derived partition as a good initialization of an efficient GP method (e.g., InfoMap) to further refine the quality of partitioning. In this setting, the online generalization and refinement of PR-GPT can not only benefit from the transfer ability regarding quality but also ensure high inference efficiency without re-training. Based on a mechanism of reducing the scale of a graph to be processed by the refinement method, PR-GPT also has the potential to support streaming GP. Experiments on the Graph Challenge benchmark demonstrate that PR-GPT can ensure faster GP on large-scale graphs without significant quality degradation, compared with running a refinement method from scratch. We will make our code public at https://github.com/KuroginQin/PRGPT.

Towards Faster Graph Partitioning via Pre-training and Inductive Inference

TL;DR

Graph partitioning is framed as a modularity-maximization problem and is typically NP-hard, motivating a scalable solution for large graphs. PR-GPT combines offline pre-training on small graphs with inductive online generalization to large graphs, followed by online refinement using fast GP refiners on a reduced, weighted super-graph, to accelerate partitioning without retraining. The approach demonstrates substantial efficiency gains on the IEEE HPEC Graph Challenge benchmark, often with negligible quality loss, and shows potential for streaming GP by progressively reducing the processing scale. The work contributes a scalable, transferable GP framework, aligns with foundation-model paradigms, and provides open-source tooling to enable efficient GP on massive graphs and streaming scenarios.

Abstract

Graph partitioning (GP) is a classic problem that divides the node set of a graph into densely-connected blocks. Following the IEEE HPEC Graph Challenge and recent advances in pre-training techniques (e.g., large-language models), we propose PR-GPT (Pre-trained & Refined Graph ParTitioning) based on a novel pre-training & refinement paradigm. We first conduct the offline pre-training of a deep graph learning (DGL) model on small synthetic graphs with various topology properties. By using the inductive inference of DGL, one can directly generalize the pre-trained model (with frozen model parameters) to large graphs and derive feasible GP results. We also use the derived partition as a good initialization of an efficient GP method (e.g., InfoMap) to further refine the quality of partitioning. In this setting, the online generalization and refinement of PR-GPT can not only benefit from the transfer ability regarding quality but also ensure high inference efficiency without re-training. Based on a mechanism of reducing the scale of a graph to be processed by the refinement method, PR-GPT also has the potential to support streaming GP. Experiments on the Graph Challenge benchmark demonstrate that PR-GPT can ensure faster GP on large-scale graphs without significant quality degradation, compared with running a refinement method from scratch. We will make our code public at https://github.com/KuroginQin/PRGPT.
Paper Structure (23 sections, 5 equations, 3 figures, 8 tables, 2 algorithms)

This paper contains 23 sections, 5 equations, 3 figures, 8 tables, 2 algorithms.

Table of Contents

  1. Introduction
  2. Problem Statement & Preliminaries
  3. Methodology
  4. Model Architecture
  5. Feature Extraction
  6. Embedding Derivation
  7. Binary Node Pair Classification
  8. Result Derivation
  9. Offline Pre-training
  10. Online Inference for Static GP
  11. Online Generalization
  12. Online Refinement
  13. Extension to Streaming GP
  14. Complexity Analysis
  15. Experiments
  16. ...and 8 more sections

Figures (3)

  • Figure 1: An overview about the applications of (i) foundation models (e.g., LLMs) and (ii) our PR-GPT method, including the (a) offline pre-training and (b) online inference. The inference of PR-GPT includes the (b.1) online generalization and (b.2) online refinement.
  • Figure 2: Model architecture and inference procedure of PR-GPT.
  • Figure 3: Evaluation results of streaming GP.