New Model Enables Cross-Domain Inference on Temporal Knowledge Graphs

Jiaxin Pan, Mojtaba Nayyeri, Osama Mohammed, Daniel Hernandez, Rongchuan Zhang, Cheng Cheng, Steffen Staab· August 12, 2026 View original

Key takeaways

  • FITTER enables cross-domain inference on temporal knowledge graphs without prior vocabulary knowledge.
  • The model uses relative time encodings and vocabulary-agnostic embeddings for flexibility.
  • It consistently outperforms inductive baselines in cross-graph transfer scenarios.
  • This approach offers a scalable foundation for reasoning over heterogeneous knowledge graphs.

Who benefits

Data AnalyticsSemantic WebAI/ML PlatformsHealthcareFinance

Summary

Researchers introduce FITTER, a novel structural model for temporal knowledge graph link prediction that allows inference across different domains, even with unseen entities and relations. It uses vocabulary-agnostic embeddings and relative time encodings to achieve cross-graph transfer without retraining.

Current methods for completing temporal knowledge graphs (TKGs) are limited because they require entities and relations to be known during training, restricting models to a single graph and vocabulary. A new model, FITTER, addresses this by offering the first fully-inductive structural approach for TKG link prediction that supports cross-domain transfer. This means it can perform inference on graphs containing entirely new entities, relation names, and timestamps from different domains. FITTER achieves this by representing predicates through their interaction patterns and encoding time based on relative rather than absolute ordering. Message-passing techniques integrate local and global temporal context to generate embeddings that are independent of specific vocabularies. The model's temporal encoding is proven to be time-shift invariant, and evaluations show it consistently outperforms existing inductive baselines in cross-domain, cross-graph transfer across various TKG benchmarks.

Why it matters

This research significantly advances the ability to leverage diverse and evolving knowledge graphs without constant retraining, making AI systems more adaptable and scalable in dynamic data environments.

How to implement this in your domain

  1. 1Explore integrating FITTER's principles into existing knowledge graph management systems for enhanced flexibility.
  2. 2Pilot cross-domain data integration projects using this approach to reduce manual schema mapping efforts.
  3. 3Evaluate the performance of vocabulary-agnostic models for real-time inference in dynamic data streams.
  4. 4Develop internal tools that can leverage relative temporal encodings for more robust time-series analysis.

Original post by Jiaxin Pan, Mojtaba Nayyeri, Osama Mohammed, Daniel Hernandez, Rongchuan Zhang, Cheng Cheng, Steffen Staab

"arXiv:2608.10668v1 Announce Type: new Abstract: Temporal knowledge graphs are central to many uses of the Semantic Web, but existing completion methods assume the entities, relation names, and timestamps to be reasoned about are already known at training time, restricting each mo…"

View on X

Originally posted by Jiaxin Pan, Mojtaba Nayyeri, Osama Mohammed, Daniel Hernandez, Rongchuan Zhang, Cheng Cheng, Steffen Staab on X · view source

Want to go deeper?

Turn these trends into skills with Learnijoy's hands-on AI & tech courses.

Explore courses

More in AI Research

AI Engineering & DevToolsAI Research

TACTICL Compresses Tabular ICL Models, Retaining Adaptability.

TACTICL is an automated framework for compressing tabular in-context learning (ICL) models by jointly pruning transformer layers and replacing them with lightweight adapters. This method significantly reduces model size and computational demands while preserving robustness to data shifts and in-context adaptability.

Mykhailo Koshil, Matthias Feurer, Katharina EggenspergerAug 12, 2026
AI Engineering & DevToolsAI Research

MoE Proxy Models Cut LLM RL Debugging Costs.

This paper introduces Mixture-of-Experts (MoE) proxy models designed for low-cost reproduction and diagnosis of failures during Large Language Model (LLM) Reinforcement Learning (RL) post-training. These proxy models significantly reduce computational resources and time needed for debugging, while accurately preserving training dynamics and fault responses.

Yikai Wang, Chuansai Zhou, Yuhang Zhou, Weiqiang Wu, Cong Wu, Yue Deng, Ben Feng, Mingming Zhu, Beirong Zhou, Zhibin Wang, Sheng Zhong, Chen Tian, Wangze ZhangAug 12, 2026
AI Engineering & DevToolsAI Research

New Algorithm Boosts Stochastic Optimal Control Efficiency.

This paper introduces Path Integral Value Matching (PI-VM), a novel value-based algorithm for Linear Quadratic Stochastic Optimal Control (LQ-SOC) that significantly improves computational efficiency and stability. By deriving a temporal recursive form of the value function and integrating Girsanov theorem with experience replay, PI-VM matches state-of-the-art precision with order-of-magnitude efficiency gains.

Bangyan Liao, Chenglei Yu, Yuchen Yang, Chuanrui Wang, Zhisheng Song, Peidong Liu, Tailin WuAug 12, 2026