Research
NVIDIA Kumo Relational Beats XGBoost on Structured Data
Hugging Face Bloghigh confidence
Why it matters
- →Challenges the dominance of gradient-boosted trees in tabular prediction tasks.
- →Eliminates the need for manual feature engineering and task-specific training.
- →Enables agentic applications to perform predictive queries directly on live databases.
NVIDIA has released Kumo Relational, a pre-trained foundation model designed for predictive modeling on relational databases. Unlike traditional approaches that require flattening tables and extensive feature engineering, the model operates directly on database schemas using graph attention mechanisms. In evaluations across 41 tasks on 15 databases, it outperformed supervised deep learning models and expert-engineered XGBoost pipelines. The system scales to databases with over 500 billion rows and is available via NVIDIA's open-source structured-data-models library or as an NIM.
Read originalThe story around this
Earlier coverage that leads up to this article, and what followed. Lines connect each piece to the closest one after it, converging here.
