FIMP: Foundation Model-Informed Message Passing for Graph Neural Networks
This work addresses the problem of integrating foundation models into graph-based tasks for researchers and practitioners in fields like biology and image analysis, representing an incremental advancement by adapting existing foundation model components to GNNs.
The paper tackled the challenge of applying foundation models to graph-structured data, which often lacks large-scale pretraining data, by proposing FIMP, a GNN framework that repurposes self-attention layers from pretrained non-textual foundation models for cross-node message-passing. It demonstrated improved performance over strong baselines on real-world image networks and biological datasets in finetuned and zero-shot settings.
Foundation models have achieved remarkable success across many domains, relying on pretraining over vast amounts of data. Graph-structured data often lacks the same scale as unstructured data, making the development of graph foundation models challenging. In this work, we propose Foundation-Informed Message Passing (FIMP), a Graph Neural Network (GNN) message-passing framework that leverages pretrained non-textual foundation models in graph-based tasks. We show that the self-attention layers of foundation models can effectively be repurposed on graphs to perform cross-node attention-based message-passing. Our model is evaluated on a real-world image network dataset and two biological applications (single-cell RNA sequencing data and fMRI brain activity recordings) in both finetuned and zero-shot settings. FIMP outperforms strong baselines, demonstrating that it can effectively leverage state-of-the-art foundation models in graph tasks.