Research map / Vision-language and multimodal / Human and embodied modeling
Relational In-Context Learning via Synthetic Pre-training with Structural Prior
Yanbo Wang, Jiaxuan You, Chuan Shi, Muhan Zhang
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add relational synthetic pre-training stage on top of single-table pre-training (ablation) | RDB-PFN-Single · single-table few-shot benchmark | accuracy | Improved |
| use RDB-PFN relational prior instead of generic single-table priors with same DFS inputs | single-estimator TFMs · 19 relational tasks | ROC-AUC | Improved |
| compare RDB-PFN to specialized single-table TFMs on single-table data | RDB-PFN · single-table benchmark | accuracy | Hurt |
| use LayerDAG learned schema generator instead of simpler DAG generator (ablation) | relational prior · 19 relational tasks | ROC-AUC | Improved |
| simplify schema generator, link mechanism, temporal component, locality assumption or interleaved attention (ablation) | RDB-PFN · 19 relational tasks | performance | Hurt |
| target-task fine-tuning beyond ICL (ablation) | RDB-PFN · relational tasks | performance | Improved |
| cross-task fine-tuning (ablation) | RDB-PFN · relational tasks | performance | Improved |
| chunked-support ensembling with larger effective support size (ablation) | RDB-PFN · relational tasks | performance | Improved |
| strengthen classical baselines with tuned XGBoost and AutoGluon best_quality (ablation) | classical baselines · relational tasks | ROC-AUC | No clear effect |
| evaluate with PR-AUC instead of ROC-AUC (ablation) | RDB-PFN comparison · imbalanced relational tasks | PR-AUC | No clear effect |
| disable default ensembling for baselines (Lite) (ablation) | tabular foundation models · 19 relational tasks | ROC-AUC | Mixed |
Closest papers by meaning
- A Pre-training Framework for Relational Data with Information-theoretic Principles (NeurIPS 2025 · similarity 0.87)
- Griffin: Towards a Graph-Centric Relational Database Foundation Model (ICML 2025 · similarity 0.87)
- Relational Learning in Pre-Trained Models: A Theory from Hypergraph Recovery Perspective (ICML 2024 · similarity 0.86)
- Joint Relational Database Generation via Graph-Conditional Diffusion Models (NeurIPS 2025 · similarity 0.85)
- RDB2G-Bench: A Comprehensive Benchmark for Automatic Graph Modeling of Relational Databases (NeurIPS 2025 · similarity 0.85)
- Understanding the Surprising Generalization Properties of Tabular Foundation Models (NeurIPS 2026 · similarity 0.84)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
