Research map / Large language models / Retrieval and knowledge
Knowledge-in-Context: Towards Knowledgeable Semi-Parametric Language Models
Xiaoman Pan, Wenlin Yao, Hongming Zhang, Dian Yu, Dong Yu, Jianshu Chen
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| Add KiC's instance-adaptive knowledge-rich memory and selector | T5-LM-adapt Large · MMLU test average | accuracy: 31.8 -> 39.4 | Improved |
| Replace English Wikipedia plain-text memory with knowledge-rich memory (ablation) | KiC Large · MMLU test average | accuracy: 33.5 -> 39.4 | Improved |
| Add a knowledge selector that routes instances to a selected knowledge category (ablation) | KiC Large · MMLU test average | accuracy: 37.4 -> 39.4 | Improved |
| Use an instance-adaptive selector instead of a task-adaptive selector (ablation) | KiC Large · MMLU test average | accuracy: 36.7 -> 39.4 | Improved |
| Add a generalist expert that can skip external knowledge (ablation) | KiC Large · MMLU test average | accuracy: 38.5 -> 39.4 | Improved |
| Add external knowledge (ablation) | KiC Large · P3 held-out tasks average | median accuracy: 51.2 -> 57.6 | Improved |
| Use knowledge-rich memory instead of English Wikipedia plain text (ablation) | KiC Large · P3 held-out tasks average | median accuracy: 53.3 -> 57.6 | Improved |
| Add a knowledge selector rather than retrieving from all categories together (ablation) | KiC Large · P3 held-out tasks average | median accuracy: 53.9 -> 57.6 | Improved |
| Use instance-adaptive rather than task-adaptive knowledge selection (ablation) | KiC Large · P3 held-out tasks average | median accuracy: 55.1 -> 57.6 | Improved |
| Add a generalist expert that can skip external knowledge (ablation) | KiC Large · P3 held-out tasks average | median accuracy: 56.6 -> 57.6 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 0 of 1 on most shared results.
- Atlas 11B — Atlas: Few-shot Learning with Retrieval Augmented Language Models (JMLR 2023) · better on 0 of 1 shared result
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Self-Updatable Large Language Models by Integrating Context into Model Parameters (ICLR 2025 · similarity 0.88)
- Seeking Neural Nuggets: Knowledge Transfer in Large Language Models from a Parametric Perspective (ICLR 2024 · similarity 0.88)
- KaSA: Knowledge-Aware Singular-Value Adaptation of Large Language Models (ICLR 2025 · similarity 0.88)
- Generative Adapter: Contextualizing Language Models in Parameters with A Single Forward Pass (ICLR 2025 · similarity 0.87)
- In-Context Editing: Learning Knowledge from Self-Induced Distributions (ICLR 2025 · similarity 0.87)
- Knowledge Fusion of Large Language Models (ICLR 2024 · similarity 0.87)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
