AI Paper Atlas

Research map / Architectures and efficiency / Attention and transformers

Understanding Task Vectors in In-Context Learning: Emergence, Functionality, and Limitations

Yuxin Dong, Jiachen Jiang, Zhihui Zhu, Xia Ning

ICLR 2026 · Architectures and efficiency › Attention and transformers · approach: Training dynamics of in-context learning · 8 citations (OpenAlex)

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 changedWhereMeasuredEffect
inject task vectors into every arrow token rather than only the final oneTask vector method on Llama-13B, 1-shot · average ICL accuracyaccuracy: 78.79 -> 79.34Improved
inject task vectors into every arrow token rather than only the final oneTask vector method on Llama-13B, 4-shot · average ICL accuracyaccuracy: 84.66 -> 85.64Improved
inject multiple task vectors on bijection tasksTask vector method on Llama-13B, 4-shot · bijection tasksaccuracy: 70.44 -> 72.53Improved
inject multiple task vectors on Translation tasksTask vector method on Llama-13B, 1-shot · Translationaccuracy: 80.67 -> 80.13No clear effect
use task vector instead of ICL on bidirectional bijection (To Upper)Llama-7B · bijection X<->Yaccuracy: 1.00 -> 0.55Hurt
use task vector instead of ICL on bidirectional English-French translationLlama-7B · bijection X<->Yaccuracy: 0.54 -> 0.35Hurt
use task vector instead of ICL on the copy taskLlama-7B · Copy (x=y)accuracy: 1.00 -> 0.98No clear effect
add one task vector to zero-shot promptsLlama-13B · average ICL accuracyaccuracy: 10.28 -> 72.26Improved
use triplet-formatted instead of pairwise prompts (ablation)linear transformer · synthetic linear regression ICL riskICL riskNo clear effect

Closest papers by meaning

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.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy