Learning to Evolve: Bayesian-Guided Continual Knowledge Graph Embedding
Linyu Li, Zhi Jin, Yuanpeng He, Dongming Jin, Yichi Zhang, Haoran Duan, Nyima Tashi · 2025
Read Learning to Evolve: Bayesian-Guided Continual Knowledge Graph Embedding by Linyu Li, Zhi Jin, Yuanpeng He, Dongming Jin, Yichi Zhang, Haoran Duan, Nyima Tashi (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2508-02426.txt.
Description
Linyu Li and co-authors teach knowledge graph embeddings to grow without forgetting. Knowledge graphs organize web facts as linked triples, and embeddings turn them into vectors models can use. Social feeds churn daily, so a frozen embedding rots, yet naive retraining erases old knowledge — the classic catastrophic forgetting. Their 2025 paper frames the update itself as Bayesian inference.
Their framework, BAKE, treats each new batch of facts as evidence for a posterior update. Yesterday's embedding becomes today's prior, so learning the new leans on the old in a principled ratio. The method aims to track memes and events as they flare without losing the stable core. Web data engineers get a memory discipline for graphs that never sit still.
Catalog record
| CREATOR | Linyu Li, Zhi Jin, Yuanpeng He, Dongming Jin, Yichi Zhang, Haoran Duan, Nyima Tashi |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 38 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2508-02426