When Prompting Meets Spiking: Graph Sparse Prompting via Spiking Graph Prompt Learning

TL;DR

This work addresses efficient adaptation of pre-trained graph neural networks by introducing Spiking Graph Prompt Feature (SpikingGPF), a sparse prompt learning framework. It combines two spiking-neuron–driven modules: (i) S-learning to obtain sparse coefficients $\\mathbf{S}$ over basis prompts $\\mathbf{B}$, and (ii) P-learning to produce sparse node prompts $\\mathbf{p}_i$, with $\\mathbf{P}=\\mathbf{S}\\mathbf{B}^T$ and $\\mathbf{p}_i$ formed from a sparse subset of $\\mathbf{B}$. Empirical results on 11 datasets show SpikingGPF outperforms standard GPF variants and other prompting methods, with improved robustness to noise and favorable performance in few-shot settings. The approach yields a lightweight, robust alternative for graph prompt tuning, enabling scalable and resilient adaptation of pre-trained GNNs to downstream tasks.

Abstract

Graph Prompt Feature (GPF) learning has been widely used in adapting pre-trained GNN model on the downstream task. GPFs first introduce some prompt atoms and then learns the optimal prompt vector for each graph node using the linear combination of prompt atoms. However, existing GPFs generally conduct prompting over node's all feature dimensions which is obviously redundant and also be sensitive to node feature noise. To overcome this issue, for the first time, this paper proposes learning sparse graph prompts by leveraging the spiking neuron mechanism, termed Spiking Graph Prompt Feature (SpikingGPF). Our approach is motivated by the observation that spiking neuron can perform inexpensive information processing and produce sparse outputs which naturally fits the task of our graph sparse prompting. Specifically, SpikingGPF has two main aspects. First, it learns a sparse prompt vector for each node by exploiting a spiking neuron architecture, enabling prompting on selective node features. This yields a more compact and lightweight prompting design while also improving robustness against node noise. Second, SpikingGPF introduces a novel prompt representation learning model based on sparse representation theory, i.e., it represents each node prompt as a sparse combination of prompt atoms. This encourages a more compact representation and also facilitates efficient computation. Extensive experiments on several benchmarks demonstrate the effectiveness and robustness of SpikingGPF.

Figures (9)

• Figure 1: Illustration of the learning architecture of our $\textbf{S}$-learning module.
• Figure 2: Visualizations of learned $\textbf{S}$ ($n=59, K=20$ in this example) on KarateClub rozemberczki2020karate dataset.
• Figure 3: Illustration of the learning architecture of our P-learning module.
• Figure 4: Visualizations of learned prompts $\textbf{P}$ on KarateClub rozemberczki2020karate dataset under different parameter $\gamma$ and $T$ values.
• Figure 5: Comparison results of different methods under Random Attack on Cora and PubMed sen2008collective datasets with perturbation rate ranging from $20\%$ to $100\%$.