Knowledge Graph Reasoning and Relational Foundation Models
Models that predict missing links and entities in multi-relational graphs or relational databases, including pretrained models that transfer to unseen graphs.
Tier Grades
Tier A: Frontier AI labs
feasible
Intel and Mila pretrain ULTRA; Kumo AI builds a relational foundation model.
Tier B: Indonesian AI labs
feasible
One pre-trained ULTRA model covers 57 graphs, so pretraining compute is modest though not stated.
Tier C: Institution units such as ITB
feasible
ULTRA came from a 5-author team, the size of an institution unit.
Tier D: Organisations such as SEMAI (PoT, PoC, PoV)
feasible
A pre-trained ULTRA model links entities zero-shot on a client's knowledge graph in a proof of concept.
Where this comes from
Tiers: a) frontier AI labs, b) Indonesian AI labs, c) institution units such as ITB, d) organisations such as SEMAI delivering proof of technology, concept and value. Each rationale cites the anchors below.
Anchors
- ComputeULTRA: a single pre-trained model; parameter count and compute are not in the abstract.
- DataULTRA: link prediction experiments on 57 different knowledge graphs, zero-shot on unseen graphs.
- Team size (proxy)Proxy only (author count of an exemplar paper, not a headcount): 5 authors on ULTRA.
- Time horizonOne to two years.
- Shelf-life signalsZero-shot transfer to unseen graphs (ULTRA on 57 KGs), Relational foundation models over enterprise databases (KumoRFM)
- Pipeline roleTier A and universities release pretrained graph reasoners; Tier C builds Indonesian knowledge graphs; Tier D uses zero-shot link prediction on client data.
Where this comes from
Team size is a proxy (author counts of exemplar papers), not a headcount.
Who and What
Links
R&D TopicKnowledge Graph Reasoning and Relational Foundation Models
Models
R-GCN (Relational Graph Convolutional Network)University of Amsterdam, graph
RotatEPeking University, graph
ULTRAIntel, graph
KumoRFMKumo AI, graph
R-GCN (Relational Graph Convolutional Network)University of Amsterdam, graphRotatEPeking University, graphULTRAIntel, graphKumoRFMKumo AI, graphUse Casesnone recorded
Who Runs Itnone recorded
Papers
- Modeling Relational Data with Graph Convolutional NetworksMichael Schlichtkrull et al., arXiv preprint, 2017 founding
- RelBench: A Benchmark for Deep Learning on Relational DatabasesJoshua Robinson et al., arXiv preprint, 2024 state of the art
- Towards Foundation Models for Knowledge Graph ReasoningMikhail Galkin et al., ICLR 2024 (per arXiv comment), 2023 state of the art
- Relational Deep Learning: Graph Representation Learning on Relational DatabasesMatthias Fey et al., arXiv preprint, 2023 state of the art
- A Survey on Knowledge Graphs: Representation, Acquisition and ApplicationsShaoxiong Ji et al., IEEE Transactions on Neural Networks and Learning Systems, 2021, 2020 survey
Where this comes from
The papers to read first: where the direction began, its current best, and a survey when one exists.
Sources
- arxiv.org/abs/2310.04562 https://arxiv.org/abs/2310.04562
To cite this page: Project AURION, AI Use-Cases Research & Innovation, "Knowledge Graph Reasoning and Relational Foundation Models", https://aurion.semai.id/topics/knowledge-graph-reasoning-and-relational-foundation-models/, records of October 2026.