Copertina

Salvato in:
Dettagli Bibliografici
Autori principali: Rizvi, Syed Asad, Pallikkavaliyaveetil, Nazreen, Zhang, David, Lyu, Zhuoyang, Nguyen, Nhi, Lyu, Haoran, Christensen, Benjamin, Caro, Josue Ortega, Fonseca, Antonio H. O., Zappala, Emanuele, Bagherian, Maryam, Averill, Christopher, Abdallah, Chadi G., Karbasi, Amin, Ying, Rex, Brbic, Maria, Dhodapkar, Rahul Madhav, van Dijk, David
Natura: Preprint
Pubblicazione: 2022
Soggetti:
Machine Learning
Accesso online:https://arxiv.org/abs/2210.09475
Tags: Aggiungi Tag
Nessun Tag, puoi essere il primo ad aggiungerne!!
Sommario:
  • 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.

Documenti analoghi