Relational foundation models like Kumo work directly on multi-table enterprise data
pandeyparul · x · 2026-11-05
Most enterprise data lives across relational tables, but tabular foundation models expect one flat table, and joins/summaries lose fine-grained information. Relational foundation models — NVIDIA's Kumo Relational and Prior Labs' TabPFN-Rel — operate directly on related tables while preserving structure, and need no per-company training since they're pre-trained. Kumo Relational uses in-context learning: it predicts from examples at inference time. The author breaks down its three-stage pipeline, starting with intra-table row embeddings.
More from Research
- TestBED: Oxford team releases open-source toolkit and large-scale benchmark for Bayesian experimental design — tom_rainforth · 2026-11-05
- NVIDIA's TAMPAgent uses factor graphs to let robots zero-shot mobile manipulation in unseen homes — danfei_xu · 2026-11-05
- Harmonic's Aristotle model synthesizes a circuit from a paper with full Lean proof — satnam6502 · 2026-11-05
- Alibaba Research's Recursa lifts DeepSeek-V4.1-Flash from 30% to 55% on legal tasks without touching weights — omarsar0 · 2026-11-05
- ByteDance releases sparse autoencoders for Qwen3-30B revealing interpretable MoE expert specializations — _akhaliq · 2026-11-05
- Cohere paper: agents find injected solutions in 80% of runs but exploit them only 37-50% of the time — burny_tech · 2026-11-05