SEPAL: Scalable Feature Learning on Huge Knowledge Graphs for Downstream Machine Learning

This repository contains the implementation of the paper:

Scalable Feature Learning on Huge Knowledge Graphs for Downstream Machine Learning
Félix Lefebvre and Gaël Varoquaux
NeurIPS 2025
PDF: https://arxiv.org/pdf/2507.00965v2

✨ Highlights

• Scales to knowledge graphs with millions of entities
• Robust to highly skewed degree distributions
• Produces embeddings for downstream regression and classification tasks

Method details and ablations are in the paper.

🧪 Example

• Mini YAGO3 tutorial: examples/mini_yago3_embeddings.ipynb

📊 Datasets

• Downstream tables and Mini YAGO3: https://huggingface.co/datasets/inria-soda/sepal-datasets

📣 Citation

If you use SEPAL, please cite:

@inproceedings{lefebvre2025scalable,
  title={Scalable Feature Learning on Huge Knowledge Graphs for Downstream Machine Learning},
  author={Lefebvre, Félix and Varoquaux, Gaël},
  booktitle={Advances in Neural Information Processing Systems},
  volume={38},
  year={2025}
}