Griffin: Towards a Graph-Centric Relational Database Foundation Model

2025-05-08

AI Summary

Existing relational database (RDB) models struggle to jointly address diverse downstream tasks. To address this limitation, we propose Griffin—the first graph-centric foundational model for RDBs. Griffin introduces a unified architecture comprising a shared data encoder and task-specific decoders, integrating heterogeneous feature encoding, enhanced message-passing neural networks (MPNNs), cross-attention mechanisms, and a novel graph aggregator. It supports both single-table and multi-table joint pretraining on large-scale real-world RDB graphs containing over 150 million nodes. Empirical evaluation shows Griffin matches or surpasses state-of-the-art (SOTA) single-task models across multiple benchmarks. Moreover, it achieves substantial accuracy gains in few-shot settings and demonstrates strong cross-dataset and cross-task transferability, significantly improving generalization under low-resource conditions.

Technology Category

• Machine Learning: Graph-based Machine Learning
• Reasoning under Uncertainty: Relational Probabilistic Models
• Data Mining & Knowledge Management: Graph Mining, Social Network Analysis & Community

Application Category

• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• Search and Retrieval-Augmented AI: Retrieval-Augmented Generation (RAG) and multi-modal RAG

Abstract

We introduce Griffin, the first foundation model attemptation designed specifically for Relational Databases (RDBs). Unlike previous smaller models focused on single RDB tasks, Griffin unifies the data encoder and task decoder to handle diverse tasks. Additionally, we enhance the architecture by incorporating a cross-attention module and a novel aggregator. Griffin utilizes pretraining on both single-table and RDB datasets, employing advanced encoders for categorical, numerical, and metadata features, along with innovative components such as cross-attention modules and enhanced message-passing neural networks (MPNNs) to capture the complexities of relational data. Evaluated on large-scale, heterogeneous, and temporal graphs extracted from RDBs across various domains (spanning over 150 million nodes), Griffin demonstrates superior or comparable performance to individually trained models, excels in low-data scenarios, and shows strong transferability with similarity and diversity in pretraining across new datasets and tasks, highlighting its potential as a universally applicable foundation model for RDBs. Code available at https://github.com/yanxwb/Griffin.

Problem

• Develops first foundation model for relational databases
• Unifies data encoder and task decoder for diverse tasks
• Enhances architecture with cross-attention and novel aggregator

Innovation

• Unified data encoder and task decoder
• Cross-attention module and novel aggregator
• Pretraining with advanced encoders and MPNNs