Relational Foundation Models
Overview
Relational foundation models aim to transfer learning across databases, where prediction depends on both table contents and relationships between tables. Common approaches include relational transformers pretrained on synthetic or real databases, in-context learning from labeled examples, and schema-aware row encoders combined with graph neural networks or fine-tuning. Benchmarks such as RelBench test prediction and transfer across databases, including how many examples models need and how well they handle large or sparse contexts. A central tension is whether pretraining truly teaches models to use cross-table structure: task scarcity, support-set leakage, and context limits can undermine transfer and make evaluation unreliable.
Latest papers
19 papers3 papers in the last four weeks, down 50% on the four weeks before. 0.0% of all new papers.