AI Paper Atlas

Research map / Architectures and efficiency / Attention and transformers

Jump Self-attention: Capturing High-order Statistics in Transformers

Haoyi Zhou, Siyang Xiao, Shanghang Zhang, Jieqi Peng, Shuai Zhang, Jianxin Li

NeurIPS 2022 · Architectures and efficiency › Attention and transformers · approach: Attention head and block redesign · 1 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
replace canonical self-attention heads with Jump Self-attentionBERT-base · GLUE devaverage score: 79.7 -> 81.7Improved
replace canonical self-attention heads with Jump Self-attentionRoBERTa-base · GLUE devaverage score: 86.4 -> 87.1Improved
add JAT to BERT-base on small datasetBERT-base · GLUE WNLIaccuracy: 53.5 -> 57.3Improved
add JAT to RoBERTa-baseRoBERTa-base · SuperGLUE devaverage score: 75.9 -> 78.5Improved
add JAT to BERT-baseBERT-base · SQuAD v2.0F1: 79.3 -> 82.0Improved
add JAT to RoBERTa-baseRoBERTa-base · SQuAD v1.1F1: 94.6 -> 95.2Improved
use JAT in layers 1-6 versus layers 7-12 (ablation)RoBERTa-base · GLUE CoLAMatthews correlation: 64.6 vs 63.8Improved
use JAT heads only in layers 9-12 (ablation)RoBERTa-base · GLUE RTEaccuracy: 78.7 -> 76.1Hurt
increase number of JAT heads from 2 to 12 in layers 1-6 (ablation)RoBERTa-JAT · GLUE RTEaccuracy: 80.2 -> 77.8Hurt
increase number of JAT heads from 2 to 12 in layers 1-6 (ablation)RoBERTa-JAT · GLUE CoLAMatthews correlation: 63.7 -> 64.7Improved
use second-order instead of first-order JAT (ablation)RoBERTa-base · GLUE RTEaccuracy: 79.9 -> 80.2Improved
use third-order instead of second-order JAT (ablation)RoBERTa-base · GLUE RTEaccuracy: 80.2 -> 78.7Hurt

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