AI Paper Atlas

Research map / Architectures and efficiency / Attention and transformers

ConvBERT: Improving BERT with Span-based Dynamic Convolution

Zi-Hang Jiang, Weihao Yu, Daquan Zhou, Yunpeng Chen, Jiashi Feng, Shuicheng Yan

NeurIPS 2020 · Architectures and efficiency › Attention and transformers · approach: Long-context KV and memory attention · 129 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 conventional self-attention with ConvBERT mixed attentionBERTbase · GLUE testAvg.: 80.9 -> 84.9Improved
use ConvBERTbase and train for 4M updatesBERTbase · GLUE testAvg.: 83.5 -> 86.4Improved
add span-based dynamic convolution to bottleneck attention (ablation)ConvBERTsmall · GLUE devGLUE: 80.6 -> 81.4Improved
add span-based dynamic convolution to bottleneck attention (ablation)ConvBERTbase · GLUE devGLUE: 84.9 -> 85.7Improved
add grouped linear feed-forward operator (ablation)ConvBERTmedium-small, BNK · GLUE devGLUE: 80.9 -> 81.0No clear effect
increase hidden and head dimensions (ablation)ConvBERTmedium-small, BNK + GL · GLUE devGLUE: 81.0 -> 81.1No clear effect
replace span-based dynamic convolution with conventional convolution (ablation)ConvBERTsmall · GLUE devAvg.: 81.4 -> 78.2Hurt
replace span-based dynamic convolution with token-conditioned dynamic convolution (ablation)ConvBERTsmall · GLUE devAvg.: 81.4 -> 80.6Hurt
add conventional depth-wise separable convolution parallel to self-attention (ablation)BERT · GLUE devAvg.Hurt
add token-conditioned dynamic convolution (ablation)BERT · GLUE devAvg.No clear effect
use mixed attention instead of self-attentionbase-sized model · inferenceSpeed (ms/sample): 17.66 -> 12.94Improved

Earlier methods this paper compares with

Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.

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

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