← Home
TWIML · May 21, 2026 · 1h 6m
Relational Foundation Models for Enterprise Data with Jure Leskovec
Jure Leskovec, co-founder of Kumo and Stanford CS professor, discusses two major research directions: AI Virtual Cell, a multiscale approach to learning representations from proteins to patients using single-cell RNA-seq and protein language models; and relational deep learning, which reframes enterprise databases as graphs for direct neural network training. He introduces Kumo's Relational Foundation Model (RFM2), demonstrating in-context learning over subgraphs for zero-shot predictions on new databases, with real-world deployments at Reddit, DoorDash, and Coinbase.
This summary was generated from show notes and public descriptions, not from a full transcript review. Details may contain inaccuracies.
Curious
•
Attention Over Tables and Columns as Explainability Mechanism00:45:00
RFM2 provides explainability by exposing attention weights across tables and columns, allowing practitioners to see which data sources and relationships drive model predictions—a transparency layer absent in many black-box ML systems.Novel
•
Relational Foundation Models for Zero-Shot Enterprise Prediction00:00:00
Kumo's Relational Foundation Model (RFM2) learns representations directly from raw multi-table database schemas and performs zero-shot predictions on new databases and tasks using in-context learning over subgraphs, eliminating the need for task-specific training data.•
AI Virtual Cell: Learning Biology Without Hand-Encoded Features00:05:00
AI Virtual Cell is a multiscale learning effort that combines single-cell RNA-seq data, protein language models (like ESM), and structure models (like AlphaFold) to learn data-driven representations from proteins to cells to patients, without requiring biologists to manually encode biological relationships.Highlights
•
Foundation models across domains (language, vision, databases) work by learning compressed representations of structural patterns in raw data, enabling transfer and few-shot learning without task-specific engineering.
Editorial
Misc
✧Leskovec's approach removes hand-encoded biology from protein-to-patient learning pipelines, letting neural networks discover structure.
✧RFM2 performs zero-shot learning on new databases without task-specific training—treating multi-table data as subgraph patterns.
✧Explainability through attention weights over tables and columns provides transparency on which data sources drive predictions.
✧Enterprise adoption already underway: Reddit, DoorDash, and Coinbase using relational foundation models in production.
Was this useful?