Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

MAGPrompt: Message-Adaptive Graph Prompt Tuning for Graph Neural Networks

Long D. Nguyen, Binh P. Nguyen

Abstract

Pre-trained graph neural networks (GNNs) transfer well, but adapting them to downstream tasks remains challenging due to mismatches between pre-training objectives and task requirements. Graph prompt tuning offers a parameter-efficient alternative to fine-tuning, yet most methods only modify inputs or representations and leave message passing unchanged, limiting their ability to adapt neighborhood interactions. We propose message-adaptive graph prompt tuning, which injects learnable prompts into the message passing step to reweight incoming neighbor messages and add task-specific prompt vectors during message aggregation, while keeping the backbone GNN frozen. The approach is compatible with common GNN backbones and pre-training strategies, and applicable across downstream settings. Experiments on diverse node- and graph-level datasets show consistent gains over prior graph prompting methods in few-shot settings, while achieving performance competitive with fine-tuning in full-shot regimes.

MAGPrompt: Message-Adaptive Graph Prompt Tuning for Graph Neural Networks

Abstract

Pre-trained graph neural networks (GNNs) transfer well, but adapting them to downstream tasks remains challenging due to mismatches between pre-training objectives and task requirements. Graph prompt tuning offers a parameter-efficient alternative to fine-tuning, yet most methods only modify inputs or representations and leave message passing unchanged, limiting their ability to adapt neighborhood interactions. We propose message-adaptive graph prompt tuning, which injects learnable prompts into the message passing step to reweight incoming neighbor messages and add task-specific prompt vectors during message aggregation, while keeping the backbone GNN frozen. The approach is compatible with common GNN backbones and pre-training strategies, and applicable across downstream settings. Experiments on diverse node- and graph-level datasets show consistent gains over prior graph prompting methods in few-shot settings, while achieving performance competitive with fine-tuning in full-shot regimes.
Paper Structure (54 sections, 28 equations, 5 figures, 12 tables)

This paper contains 54 sections, 28 equations, 5 figures, 12 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Graph Pre-training
  4. Graph Prompt Learning
  5. Preliminaries
  6. Graphs and Notation.
  7. Message Passing in GNNs.
  8. Pre-training and Fine-tuning.
  9. Graph Prompt Tuning.
  10. Scope of This Work.
  11. Method
  12. Overview
  13. MAGPrompt
  14. Message-adaptive reweighting.
  15. Prompted combination.
  16. ...and 39 more sections

Figures (5)

  • Figure 1: Illustration of MAGPrompt. (a) Input graph. (b) Standard message passing aggregates messages from neighbors of $v_1$. (c) MAGPrompt introduces edge-specific message-adaptive modulation $f(\mathbf{m}_{ij}) = a_{ij} \mathbf{m}_{ij} + \mathbf{p}_{ij}$ before aggregation.
  • Figure 2: Performance of MAGPrompt+ across different numbers of prompt bases $M_l$ on ENZYMES, DD and NCI1 datasets.
  • Figure 3: Performance of MAGPrompt+ across different $\lambda_{\mathrm{pc}}$ across ENZYMES, NCI1 and NCI109 datasets.
  • Figure 4: Performance of MAGPrompt+ across different $d_a$ on ENZYMES dataset.
  • Figure 5: Evolution of prompt mixture weights during training without (left) and with (right) the prompt-collapse regularization $\mathcal{L}_{\mathrm{pc}}$. Each stacked area shows the mean contribution of a prompt basis across epochs, while the dotted line indicates the uniform baseline. Without $\mathcal{L}_{\mathrm{pc}}$, the prompt distribution collapses toward a small subset of prompts, whereas $\mathcal{L}_{\mathrm{pc}}$ encourages balanced and stable utilization of all prompt bases.