AI Paper Atlas

Research map / Architectures and efficiency / Attention and transformers

How Transformers Learn Causal Structures In-Context: Explainable Mechanism Meets Theoretical Guarantee

Jianzhe Wei, Siyu Chen, Jianliang He, Zhuoran Yang

ICLR 2026 · Architectures and efficiency › Attention and transformers · approach: Training dynamics of in-context learning

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
train the transformer for more steps on in-context causal structure learning2-layer transformer with RPE · Parent selection loss vs BMAparent selection lossImproved
use smaller training length L (ablation)2-layer transformer with RPE · Generalization over test sample sizes L'parent selection loss gap to BMAImproved
increase test sample size L' (ablation)trained transformer · Parent selection generalizationparent selection lossImproved
compare trained weights to BMA parameter via row softmaxtrained transformer Wtf · Parameter alignment (d=20, H=50, L=3)softmax error: 0.350414 -> 0.027767Mixed

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